4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Whamcloud, Inc.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/osd/osd_handler.c
38 * Top-level entry points into osd module
40 * Author: Nikita Danilov <nikita@clusterfs.com>
41 * Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
44 #define DEBUG_SUBSYSTEM S_MDS
46 #include <linux/module.h>
48 /* LUSTRE_VERSION_CODE */
49 #include <lustre_ver.h>
50 /* prerequisite for linux/xattr.h */
51 #include <linux/types.h>
52 /* prerequisite for linux/xattr.h */
54 /* XATTR_{REPLACE,CREATE} */
55 #include <linux/xattr.h>
60 * struct OBD_{ALLOC,FREE}*()
63 #include <obd_support.h>
64 /* struct ptlrpc_thread */
65 #include <lustre_net.h>
68 #include <lustre_fid.h>
70 #include "osd_internal.h"
73 /* llo_* api support */
74 #include <md_object.h>
75 /* dt_acct_features */
78 #ifdef HAVE_LDISKFS_PDO
80 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
81 "ldiskfs with parallel directory operations");
86 static const char dot[] = ".";
87 static const char dotdot[] = "..";
88 static const char remote_obj_dir[] = "REM_OBJ_DIR";
90 static const struct lu_object_operations osd_lu_obj_ops;
91 static const struct dt_object_operations osd_obj_ops;
92 static const struct dt_object_operations osd_obj_ea_ops;
93 static const struct dt_object_operations osd_obj_otable_it_ops;
94 static const struct dt_index_operations osd_index_iam_ops;
95 static const struct dt_index_operations osd_index_ea_ops;
97 static int osd_has_index(const struct osd_object *obj)
99 return obj->oo_dt.do_index_ops != NULL;
102 static int osd_object_invariant(const struct lu_object *l)
104 return osd_invariant(osd_obj(l));
107 #ifdef HAVE_QUOTA_SUPPORT
109 osd_push_ctxt(const struct lu_env *env, struct osd_ctxt *save, bool is_md)
122 save->oc_uid = current_fsuid();
123 save->oc_gid = current_fsgid();
124 save->oc_cap = current_cap();
125 if ((tc = prepare_creds())) {
126 tc->fsuid = uc->mu_fsuid;
127 tc->fsgid = uc->mu_fsgid;
130 /* XXX not suboptimal */
131 cfs_curproc_cap_unpack(uc->mu_cap);
135 osd_pop_ctxt(struct osd_ctxt *save, bool is_md)
143 if ((tc = prepare_creds())) {
144 tc->fsuid = save->oc_uid;
145 tc->fsgid = save->oc_gid;
146 tc->cap_effective = save->oc_cap;
153 * Concurrency: doesn't matter
155 static int osd_read_locked(const struct lu_env *env, struct osd_object *o)
157 return osd_oti_get(env)->oti_r_locks > 0;
161 * Concurrency: doesn't matter
163 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
165 struct osd_thread_info *oti = osd_oti_get(env);
166 return oti->oti_w_locks > 0 && o->oo_owner == env;
170 * Concurrency: doesn't access mutable data
172 static int osd_root_get(const struct lu_env *env,
173 struct dt_device *dev, struct lu_fid *f)
175 lu_local_obj_fid(f, OSD_FS_ROOT_OID);
180 * OSD object methods.
184 * Concurrency: no concurrent access is possible that early in object
187 static struct lu_object *osd_object_alloc(const struct lu_env *env,
188 const struct lu_object_header *hdr,
191 struct osd_object *mo;
197 l = &mo->oo_dt.do_lu;
198 dt_object_init(&mo->oo_dt, NULL, d);
199 if (osd_dev(d)->od_iop_mode)
200 mo->oo_dt.do_ops = &osd_obj_ea_ops;
202 mo->oo_dt.do_ops = &osd_obj_ops;
204 l->lo_ops = &osd_lu_obj_ops;
205 cfs_init_rwsem(&mo->oo_sem);
206 cfs_init_rwsem(&mo->oo_ext_idx_sem);
207 cfs_spin_lock_init(&mo->oo_guard);
214 static int osd_get_lma(struct inode *inode, struct dentry *dentry,
215 struct lustre_mdt_attrs *lma)
219 dentry->d_inode = inode;
220 rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA, (void *)lma,
223 /* Check LMA compatibility */
224 if (lma->lma_incompat & ~cpu_to_le32(LMA_INCOMPAT_SUPP)) {
225 CWARN("%.16s: unsupported incompat LMA feature(s) "
227 LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
228 inode->i_ino, le32_to_cpu(lma->lma_incompat) &
232 lustre_lma_swab(lma);
235 } else if (rc == 0) {
243 * retrieve object from backend ext fs.
245 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
246 struct osd_inode_id *id)
248 struct inode *inode = NULL;
250 inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
252 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
253 id->oii_ino, PTR_ERR(inode));
254 } else if (id->oii_gen != OSD_OII_NOGEN &&
255 inode->i_generation != id->oii_gen) {
256 CDEBUG(D_INODE, "unmatched inode: ino = %u, gen0 = %u, "
258 id->oii_ino, id->oii_gen, inode->i_generation);
260 inode = ERR_PTR(-ESTALE);
261 } else if (inode->i_nlink == 0) {
262 /* due to parallel readdir and unlink,
263 * we can have dead inode here. */
264 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
265 make_bad_inode(inode);
267 inode = ERR_PTR(-ESTALE);
268 } else if (is_bad_inode(inode)) {
269 CWARN("%.16s: bad inode: ino = %u\n",
270 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
272 inode = ERR_PTR(-ENOENT);
274 if (id->oii_gen == OSD_OII_NOGEN)
275 osd_id_gen(id, inode->i_ino, inode->i_generation);
277 /* Do not update file c/mtime in ldiskfs.
278 * NB: we don't have any lock to protect this because we don't
279 * have reference on osd_object now, but contention with
280 * another lookup + attr_set can't happen in the tiny window
281 * between if (...) and set S_NOCMTIME. */
282 if (!(inode->i_flags & S_NOCMTIME))
283 inode->i_flags |= S_NOCMTIME;
288 struct inode *osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
289 struct osd_inode_id *id, struct lu_fid *fid)
291 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
295 inode = osd_iget(info, dev, id);
299 rc = osd_get_lma(inode, &info->oti_obj_dentry, lma);
301 *fid = lma->lma_self_fid;
302 } else if (rc == -ENODATA) {
303 LU_IGIF_BUILD(fid, inode->i_ino, inode->i_generation);
311 static struct inode *
312 osd_iget_verify(struct osd_thread_info *info, struct osd_device *dev,
313 struct osd_inode_id *id, const struct lu_fid *fid)
315 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
319 inode = osd_iget(info, dev, id);
323 rc = osd_get_lma(inode, &info->oti_obj_dentry, lma);
325 if (rc == -ENODATA) {
326 CDEBUG(D_LFSCK, "inconsistent obj: NULL, %lu, "DFID"\n",
327 inode->i_ino, PFID(fid));
334 if (!lu_fid_eq(fid, &lma->lma_self_fid)) {
335 CDEBUG(D_LFSCK, "inconsistent obj: "DFID", %lu, "DFID"\n",
336 PFID(&lma->lma_self_fid), inode->i_ino, PFID(fid));
338 return ERR_PTR(EREMCHG);
343 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
344 const struct lu_fid *fid,
345 const struct lu_object_conf *conf)
347 struct osd_thread_info *info;
348 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
349 struct osd_device *dev;
350 struct osd_idmap_cache *oic;
351 struct osd_inode_id *id;
353 struct osd_scrub *scrub;
354 struct scrub_file *sf;
359 LINVRNT(osd_invariant(obj));
360 LASSERT(obj->oo_inode == NULL);
361 LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
364 scrub = &dev->od_scrub;
365 sf = &scrub->os_file;
366 info = osd_oti_get(env);
368 oic = &info->oti_cache;
371 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
374 if (fid_is_norm(fid)) {
375 /* Search order: 1. per-thread cache. */
376 if (lu_fid_eq(fid, &oic->oic_fid)) {
378 } else if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
379 /* Search order: 2. OI scrub pending list. */
380 result = osd_oii_lookup(dev, fid, id);
385 if (sf->sf_flags & SF_INCONSISTENT)
390 * Objects are created as locking anchors or place holders for objects
391 * yet to be created. No need to osd_oi_lookup() at here because FID
392 * shouldn't never be re-used, if it's really a duplicate FID from
393 * unexpected reason, we should be able to detect it later by calling
394 * do_create->osd_oi_insert()
396 if (conf != NULL && (conf->loc_flags & LOC_F_NEW) != 0)
397 GOTO(out, result = 0);
399 /* Search order: 3. OI files. */
400 result = osd_oi_lookup(info, dev, fid, id);
401 if (result == -ENOENT) {
402 if (!fid_is_norm(fid) ||
403 !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
405 GOTO(out, result = 0);
415 inode = osd_iget(info, dev, id);
417 inode = osd_iget_verify(info, dev, id, fid);
419 result = PTR_ERR(inode);
420 if (result == -ENOENT || result == -ESTALE) {
421 fid_zero(&oic->oic_fid);
423 } else if (result == -EREMCHG) {
426 if (thread_is_running(&scrub->os_thread)) {
427 result = -EINPROGRESS;
428 } else if (!scrub->os_no_scrub) {
429 result = osd_scrub_start(dev);
430 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC "
431 "for "DFID", rc = %d [1]\n",
432 LDISKFS_SB(osd_sb(dev))->s_es->\
433 s_volume_name,PFID(fid), result);
434 if (result == 0 || result == -EALREADY)
435 result = -EINPROGRESS;
444 obj->oo_inode = inode;
445 LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
446 if (dev->od_iop_mode) {
447 obj->oo_compat_dot_created = 1;
448 obj->oo_compat_dotdot_created = 1;
451 if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
452 GOTO(out, result = 0);
454 LASSERT(obj->oo_hl_head == NULL);
455 obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
456 if (obj->oo_hl_head == NULL) {
457 obj->oo_inode = NULL;
459 GOTO(out, result = -ENOMEM);
461 GOTO(out, result = 0);
464 LINVRNT(osd_invariant(obj));
469 * Concurrency: shouldn't matter.
471 static void osd_object_init0(struct osd_object *obj)
473 LASSERT(obj->oo_inode != NULL);
474 obj->oo_dt.do_body_ops = &osd_body_ops;
475 obj->oo_dt.do_lu.lo_header->loh_attr |=
476 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
480 * Concurrency: no concurrent access is possible that early in object
483 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
484 const struct lu_object_conf *conf)
486 struct osd_object *obj = osd_obj(l);
489 LINVRNT(osd_invariant(obj));
491 result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
492 obj->oo_dt.do_body_ops = &osd_body_ops_new;
494 if (obj->oo_inode != NULL) {
495 osd_object_init0(obj);
496 } else if (fid_is_otable_it(&l->lo_header->loh_fid)) {
497 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
498 /* LFSCK iterator object is special without inode */
499 l->lo_header->loh_attr |= LOHA_EXISTS;
502 LINVRNT(osd_invariant(obj));
507 * Concurrency: no concurrent access is possible that late in object
510 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
512 struct osd_object *obj = osd_obj(l);
514 LINVRNT(osd_invariant(obj));
516 dt_object_fini(&obj->oo_dt);
517 if (obj->oo_hl_head != NULL)
518 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
523 * Concurrency: no concurrent access is possible that late in object
526 static void osd_index_fini(struct osd_object *o)
528 struct iam_container *bag;
530 if (o->oo_dir != NULL) {
531 bag = &o->oo_dir->od_container;
532 if (o->oo_inode != NULL) {
533 if (bag->ic_object == o->oo_inode)
534 iam_container_fini(bag);
536 OBD_FREE_PTR(o->oo_dir);
542 * Concurrency: no concurrent access is possible that late in object
543 * life-cycle (for all existing callers, that is. New callers have to provide
544 * their own locking.)
546 static int osd_inode_unlinked(const struct inode *inode)
548 return inode->i_nlink == 0;
552 OSD_TXN_OI_DELETE_CREDITS = 20,
553 OSD_TXN_INODE_DELETE_CREDITS = 20
560 #if OSD_THANDLE_STATS
562 * Set time when the handle is allocated
564 static void osd_th_alloced(struct osd_thandle *oth)
566 oth->oth_alloced = cfs_time_current();
570 * Set time when the handle started
572 static void osd_th_started(struct osd_thandle *oth)
574 oth->oth_started = cfs_time_current();
578 * Helper function to convert time interval to microseconds packed in
579 * long int (default time units for the counter in "stats" initialized
580 * by lu_time_init() )
582 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
586 cfs_duration_usec(cfs_time_sub(end, start), &val);
587 return val.tv_sec * 1000000 + val.tv_usec;
591 * Check whether the we deal with this handle for too long.
593 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
594 cfs_time_t alloced, cfs_time_t started,
597 cfs_time_t now = cfs_time_current();
599 LASSERT(dev != NULL);
601 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
602 interval_to_usec(alloced, started));
603 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
604 interval_to_usec(started, closed));
605 lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
606 interval_to_usec(closed, now));
608 if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
609 CWARN("transaction handle %p was open for too long: "
611 "alloced "CFS_TIME_T" ,"
612 "started "CFS_TIME_T" ,"
613 "closed "CFS_TIME_T"\n",
614 oth, now, alloced, started, closed);
615 libcfs_debug_dumpstack(NULL);
619 #define OSD_CHECK_SLOW_TH(oth, dev, expr) \
621 cfs_time_t __closed = cfs_time_current(); \
622 cfs_time_t __alloced = oth->oth_alloced; \
623 cfs_time_t __started = oth->oth_started; \
626 __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \
629 #else /* OSD_THANDLE_STATS */
631 #define osd_th_alloced(h) do {} while(0)
632 #define osd_th_started(h) do {} while(0)
633 #define OSD_CHECK_SLOW_TH(oth, dev, expr) expr
635 #endif /* OSD_THANDLE_STATS */
638 * Concurrency: doesn't access mutable data.
640 static int osd_param_is_sane(const struct osd_device *dev,
641 const struct thandle *th)
643 struct osd_thandle *oh;
644 oh = container_of0(th, struct osd_thandle, ot_super);
645 return oh->ot_credits <= osd_journal(dev)->j_max_transaction_buffers;
649 * Concurrency: shouldn't matter.
651 #ifdef HAVE_LDISKFS_JOURNAL_CALLBACK_ADD
652 static void osd_trans_commit_cb(struct super_block *sb,
653 struct journal_callback *jcb, int error)
655 static void osd_trans_commit_cb(struct journal_callback *jcb, int error)
658 struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
659 struct thandle *th = &oh->ot_super;
660 struct lu_device *lud = &th->th_dev->dd_lu_dev;
661 struct dt_txn_commit_cb *dcb, *tmp;
663 LASSERT(oh->ot_handle == NULL);
666 CERROR("transaction @0x%p commit error: %d\n", th, error);
668 dt_txn_hook_commit(th);
670 /* call per-transaction callbacks if any */
671 cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
672 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
673 "commit callback entry: magic=%x name='%s'\n",
674 dcb->dcb_magic, dcb->dcb_name);
675 cfs_list_del_init(&dcb->dcb_linkage);
676 dcb->dcb_func(NULL, th, dcb, error);
679 lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th);
683 lu_context_exit(&th->th_ctx);
684 lu_context_fini(&th->th_ctx);
688 static struct thandle *osd_trans_create(const struct lu_env *env,
691 struct osd_thread_info *oti = osd_oti_get(env);
692 struct osd_iobuf *iobuf = &oti->oti_iobuf;
693 struct osd_thandle *oh;
697 /* on pending IO in this thread should left from prev. request */
698 LASSERT(cfs_atomic_read(&iobuf->dr_numreqs) == 0);
700 th = ERR_PTR(-ENOMEM);
701 OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
703 oh->ot_quota_trans = &oti->oti_quota_trans;
704 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
708 th->th_tags = LCT_TX_HANDLE;
710 oti->oti_dev = osd_dt_dev(d);
711 CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
718 * Concurrency: shouldn't matter.
720 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
723 struct osd_thread_info *oti = osd_oti_get(env);
724 struct osd_device *dev = osd_dt_dev(d);
726 struct osd_thandle *oh;
731 LASSERT(current->journal_info == NULL);
733 oh = container_of0(th, struct osd_thandle, ot_super);
735 LASSERT(oh->ot_handle == NULL);
737 rc = dt_txn_hook_start(env, d, th);
741 if (!osd_param_is_sane(dev, th)) {
742 CWARN("%.16s: too many transaction credits (%d > %d)\n",
743 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
745 osd_journal(dev)->j_max_transaction_buffers);
746 /* XXX Limit the credits to 'max_transaction_buffers', and
747 * let the underlying filesystem to catch the error if
748 * we really need so many credits.
750 * This should be removed when we can calculate the
751 * credits precisely. */
752 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
753 #ifdef OSD_TRACK_DECLARES
754 CERROR(" attr_set: %d, punch: %d, xattr_set: %d,\n",
755 oh->ot_declare_attr_set, oh->ot_declare_punch,
756 oh->ot_declare_xattr_set);
757 CERROR(" create: %d, ref_add: %d, ref_del: %d, write: %d\n",
758 oh->ot_declare_create, oh->ot_declare_ref_add,
759 oh->ot_declare_ref_del, oh->ot_declare_write);
760 CERROR(" insert: %d, delete: %d, destroy: %d\n",
761 oh->ot_declare_insert, oh->ot_declare_delete,
762 oh->ot_declare_destroy);
767 * XXX temporary stuff. Some abstraction layer should
770 jh = ldiskfs_journal_start_sb(osd_sb(dev), oh->ot_credits);
774 LASSERT(oti->oti_txns == 0);
775 lu_context_init(&th->th_ctx, th->th_tags);
776 lu_context_enter(&th->th_ctx);
778 lu_device_get(&d->dd_lu_dev);
779 oh->ot_dev_link = lu_ref_add(&d->dd_lu_dev.ld_reference,
783 * XXX: current rule is that we first start tx,
784 * then lock object(s), but we can't use
785 * this rule for data (due to locking specifics
786 * in ldiskfs). also in long-term we'd like to
787 * use usually-used (locks;tx) ordering. so,
788 * UGLY thing is that we'll use one ordering for
789 * data (ofd) and reverse ordering for metadata
790 * (mdd). then at some point we'll fix the latter
793 LASSERT(oti->oti_r_locks == 0);
794 LASSERT(oti->oti_w_locks == 0);
807 * Concurrency: shouldn't matter.
809 static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
812 struct osd_thandle *oh;
813 struct osd_thread_info *oti = osd_oti_get(env);
814 struct osd_iobuf *iobuf = &oti->oti_iobuf;
815 struct qsd_instance *qsd = oti->oti_dev->od_quota_slave;
818 oh = container_of0(th, struct osd_thandle, ot_super);
821 /* inform the quota slave device that the transaction is
823 qsd_op_end(env, qsd, oh->ot_quota_trans);
824 oh->ot_quota_trans = NULL;
826 if (oh->ot_handle != NULL) {
827 handle_t *hdl = oh->ot_handle;
830 * add commit callback
831 * notice we don't do this in osd_trans_start()
832 * as underlying transaction can change during truncate
834 osd_journal_callback_set(hdl, osd_trans_commit_cb,
837 LASSERT(oti->oti_txns == 1);
840 * XXX: current rule is that we first start tx,
841 * then lock object(s), but we can't use
842 * this rule for data (due to locking specifics
843 * in ldiskfs). also in long-term we'd like to
844 * use usually-used (locks;tx) ordering. so,
845 * UGLY thing is that we'll use one ordering for
846 * data (ofd) and reverse ordering for metadata
847 * (mdd). then at some point we'll fix the latter
849 if (osd_dt_dev(th->th_dev)->od_is_md) {
850 LASSERT(oti->oti_r_locks == 0);
851 LASSERT(oti->oti_w_locks == 0);
853 rc = dt_txn_hook_stop(env, th);
855 CERROR("Failure in transaction hook: %d\n", rc);
857 /* hook functions might modify th_sync */
858 hdl->h_sync = th->th_sync;
860 oh->ot_handle = NULL;
861 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
862 rc = ldiskfs_journal_stop(hdl));
864 CERROR("Failure to stop transaction: %d\n", rc);
869 /* as we want IO to journal and data IO be concurrent, we don't block
870 * awaiting data IO completion in osd_do_bio(), instead we wait here
871 * once transaction is submitted to the journal. all reqular requests
872 * don't do direct IO (except read/write), thus this wait_event becomes
875 * IMPORTANT: we have to wait till any IO submited by the thread is
876 * completed otherwise iobuf may be corrupted by different request
878 cfs_wait_event(iobuf->dr_wait,
879 cfs_atomic_read(&iobuf->dr_numreqs) == 0);
881 rc = iobuf->dr_error;
886 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
888 struct osd_thandle *oh = container_of0(th, struct osd_thandle,
891 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
892 LASSERT(&dcb->dcb_func != NULL);
893 cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
899 * Called just before object is freed. Releases all resources except for
900 * object itself (that is released by osd_object_free()).
902 * Concurrency: no concurrent access is possible that late in object
905 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
907 struct osd_object *obj = osd_obj(l);
908 struct inode *inode = obj->oo_inode;
910 LINVRNT(osd_invariant(obj));
913 * If object is unlinked remove fid->ino mapping from object index.
919 obj->oo_inode = NULL;
924 * Concurrency: ->loo_object_release() is called under site spin-lock.
926 static void osd_object_release(const struct lu_env *env,
932 * Concurrency: shouldn't matter.
934 static int osd_object_print(const struct lu_env *env, void *cookie,
935 lu_printer_t p, const struct lu_object *l)
937 struct osd_object *o = osd_obj(l);
940 if (o->oo_dir != NULL)
941 d = o->oo_dir->od_container.ic_descr;
944 return (*p)(env, cookie,
945 LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
947 o->oo_inode ? o->oo_inode->i_ino : 0UL,
948 o->oo_inode ? o->oo_inode->i_generation : 0,
949 d ? d->id_ops->id_name : "plain");
953 * Concurrency: shouldn't matter.
955 int osd_statfs(const struct lu_env *env, struct dt_device *d,
956 struct obd_statfs *sfs)
958 struct osd_device *osd = osd_dt_dev(d);
959 struct super_block *sb = osd_sb(osd);
960 struct kstatfs *ksfs;
963 if (unlikely(osd->od_mnt == NULL))
966 /* osd_lproc.c call this without env, allocate ksfs for that case */
967 if (unlikely(env == NULL)) {
972 ksfs = &osd_oti_get(env)->oti_ksfs;
975 cfs_spin_lock(&osd->od_osfs_lock);
977 if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
978 result = sb->s_op->statfs(sb->s_root, ksfs);
979 if (likely(result == 0)) { /* N.B. statfs can't really fail */
980 osd->od_osfs_age = cfs_time_current_64();
981 statfs_pack(&osd->od_statfs, ksfs);
985 if (likely(result == 0))
986 *sfs = osd->od_statfs;
987 cfs_spin_unlock(&osd->od_osfs_lock);
989 if (unlikely(env == NULL))
996 * Concurrency: doesn't access mutable data.
998 static void osd_conf_get(const struct lu_env *env,
999 const struct dt_device *dev,
1000 struct dt_device_param *param)
1002 struct super_block *sb = osd_sb(osd_dt_dev(dev));
1005 * XXX should be taken from not-yet-existing fs abstraction layer.
1007 param->ddp_mnt = osd_dt_dev(dev)->od_mnt;
1008 param->ddp_max_name_len = LDISKFS_NAME_LEN;
1009 param->ddp_max_nlink = LDISKFS_LINK_MAX;
1010 param->ddp_block_shift = sb->s_blocksize_bits;
1011 param->ddp_mount_type = LDD_MT_LDISKFS;
1012 param->ddp_mntopts = 0;
1013 if (test_opt(sb, XATTR_USER))
1014 param->ddp_mntopts |= MNTOPT_USERXATTR;
1015 if (test_opt(sb, POSIX_ACL))
1016 param->ddp_mntopts |= MNTOPT_ACL;
1018 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1019 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1020 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE;
1023 param->ddp_max_ea_size = sb->s_blocksize;
1028 * Helper function to get and fill the buffer with input values.
1030 static struct lu_buf *osd_buf_get(const struct lu_env *env, void *area, ssize_t len)
1034 buf = &osd_oti_get(env)->oti_buf;
1041 * Concurrency: shouldn't matter.
1043 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1045 CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1046 return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1050 * Start commit for OSD device.
1052 * An implementation of dt_commit_async method for OSD device.
1053 * Asychronously starts underlayng fs sync and thereby a transaction
1056 * \param env environment
1057 * \param d dt device
1059 * \see dt_device_operations
1061 static int osd_commit_async(const struct lu_env *env,
1062 struct dt_device *d)
1064 struct super_block *s = osd_sb(osd_dt_dev(d));
1067 CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1068 RETURN(s->s_op->sync_fs(s, 0));
1072 * Concurrency: shouldn't matter.
1075 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1077 struct super_block *sb = osd_sb(osd_dt_dev(d));
1081 CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1083 rc = __lvfs_set_rdonly(sb->s_bdev, LDISKFS_SB(sb)->journal_bdev);
1088 * Concurrency: serialization provided by callers.
1090 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1091 int mode, unsigned long timeout, __u32 alg,
1092 struct lustre_capa_key *keys)
1094 struct osd_device *dev = osd_dt_dev(d);
1097 dev->od_fl_capa = mode;
1098 dev->od_capa_timeout = timeout;
1099 dev->od_capa_alg = alg;
1100 dev->od_capa_keys = keys;
1105 * Concurrency: serialization provided by callers.
1107 static void osd_init_quota_ctxt(const struct lu_env *env, struct dt_device *d,
1108 struct dt_quota_ctxt *ctxt, void *data)
1110 struct obd_device *obd = (void *)ctxt;
1111 struct vfsmount *mnt = (struct vfsmount *)data;
1114 obd->u.obt.obt_sb = mnt->mnt_root->d_inode->i_sb;
1115 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1116 obd->obd_lvfs_ctxt.pwdmnt = mnt;
1117 obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
1118 obd->obd_lvfs_ctxt.fs = get_ds();
1124 * Note: we do not count into QUOTA here.
1125 * If we mount with --data_journal we may need more.
1127 const int osd_dto_credits_noquota[DTO_NR] = {
1130 * INDEX_EXTRA_TRANS_BLOCKS(8) +
1131 * SINGLEDATA_TRANS_BLOCKS(8)
1132 * XXX Note: maybe iam need more, since iam have more level than
1135 [DTO_INDEX_INSERT] = 16,
1136 [DTO_INDEX_DELETE] = 16,
1140 [DTO_INDEX_UPDATE] = 16,
1142 * Create a object. The same as create object in EXT3.
1143 * DATA_TRANS_BLOCKS(14) +
1144 * INDEX_EXTRA_BLOCKS(8) +
1145 * 3(inode bits, groups, GDT)
1147 [DTO_OBJECT_CREATE] = 25,
1149 * XXX: real credits to be fixed
1151 [DTO_OBJECT_DELETE] = 25,
1153 * Attr set credits (inode)
1155 [DTO_ATTR_SET_BASE] = 1,
1157 * Xattr set. The same as xattr of EXT3.
1158 * DATA_TRANS_BLOCKS(14)
1159 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1160 * are also counted in. Do not know why?
1162 [DTO_XATTR_SET] = 14,
1165 * credits for inode change during write.
1167 [DTO_WRITE_BASE] = 3,
1169 * credits for single block write.
1171 [DTO_WRITE_BLOCK] = 14,
1173 * Attr set credits for chown.
1174 * This is extra credits for setattr, and it is null without quota
1176 [DTO_ATTR_SET_CHOWN]= 0
1179 static const struct dt_device_operations osd_dt_ops = {
1180 .dt_root_get = osd_root_get,
1181 .dt_statfs = osd_statfs,
1182 .dt_trans_create = osd_trans_create,
1183 .dt_trans_start = osd_trans_start,
1184 .dt_trans_stop = osd_trans_stop,
1185 .dt_trans_cb_add = osd_trans_cb_add,
1186 .dt_conf_get = osd_conf_get,
1187 .dt_sync = osd_sync,
1189 .dt_commit_async = osd_commit_async,
1190 .dt_init_capa_ctxt = osd_init_capa_ctxt,
1191 .dt_init_quota_ctxt= osd_init_quota_ctxt,
1194 static void osd_object_read_lock(const struct lu_env *env,
1195 struct dt_object *dt, unsigned role)
1197 struct osd_object *obj = osd_dt_obj(dt);
1198 struct osd_thread_info *oti = osd_oti_get(env);
1200 LINVRNT(osd_invariant(obj));
1202 LASSERT(obj->oo_owner != env);
1203 cfs_down_read_nested(&obj->oo_sem, role);
1205 LASSERT(obj->oo_owner == NULL);
1209 static void osd_object_write_lock(const struct lu_env *env,
1210 struct dt_object *dt, unsigned role)
1212 struct osd_object *obj = osd_dt_obj(dt);
1213 struct osd_thread_info *oti = osd_oti_get(env);
1215 LINVRNT(osd_invariant(obj));
1217 LASSERT(obj->oo_owner != env);
1218 cfs_down_write_nested(&obj->oo_sem, role);
1220 LASSERT(obj->oo_owner == NULL);
1221 obj->oo_owner = env;
1225 static void osd_object_read_unlock(const struct lu_env *env,
1226 struct dt_object *dt)
1228 struct osd_object *obj = osd_dt_obj(dt);
1229 struct osd_thread_info *oti = osd_oti_get(env);
1231 LINVRNT(osd_invariant(obj));
1233 LASSERT(oti->oti_r_locks > 0);
1235 cfs_up_read(&obj->oo_sem);
1238 static void osd_object_write_unlock(const struct lu_env *env,
1239 struct dt_object *dt)
1241 struct osd_object *obj = osd_dt_obj(dt);
1242 struct osd_thread_info *oti = osd_oti_get(env);
1244 LINVRNT(osd_invariant(obj));
1246 LASSERT(obj->oo_owner == env);
1247 LASSERT(oti->oti_w_locks > 0);
1249 obj->oo_owner = NULL;
1250 cfs_up_write(&obj->oo_sem);
1253 static int osd_object_write_locked(const struct lu_env *env,
1254 struct dt_object *dt)
1256 struct osd_object *obj = osd_dt_obj(dt);
1258 LINVRNT(osd_invariant(obj));
1260 return obj->oo_owner == env;
1263 static int capa_is_sane(const struct lu_env *env,
1264 struct osd_device *dev,
1265 struct lustre_capa *capa,
1266 struct lustre_capa_key *keys)
1268 struct osd_thread_info *oti = osd_oti_get(env);
1269 struct lustre_capa *tcapa = &oti->oti_capa;
1270 struct obd_capa *oc;
1274 oc = capa_lookup(dev->od_capa_hash, capa, 0);
1276 if (capa_is_expired(oc)) {
1277 DEBUG_CAPA(D_ERROR, capa, "expired");
1284 if (capa_is_expired_sec(capa)) {
1285 DEBUG_CAPA(D_ERROR, capa, "expired");
1289 cfs_spin_lock(&capa_lock);
1290 for (i = 0; i < 2; i++) {
1291 if (keys[i].lk_keyid == capa->lc_keyid) {
1292 oti->oti_capa_key = keys[i];
1296 cfs_spin_unlock(&capa_lock);
1299 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1303 rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1307 if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1308 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1312 oc = capa_add(dev->od_capa_hash, capa);
1318 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1319 struct lustre_capa *capa, __u64 opc)
1321 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1322 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1323 struct md_capainfo *ci;
1326 if (!dev->od_fl_capa)
1329 if (capa == BYPASS_CAPA)
1332 ci = md_capainfo(env);
1336 if (ci->mc_auth == LC_ID_NONE)
1340 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1344 if (!lu_fid_eq(fid, &capa->lc_fid)) {
1345 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1350 if (!capa_opc_supported(capa, opc)) {
1351 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1355 if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1356 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1363 static struct timespec *osd_inode_time(const struct lu_env *env,
1364 struct inode *inode, __u64 seconds)
1366 struct osd_thread_info *oti = osd_oti_get(env);
1367 struct timespec *t = &oti->oti_time;
1369 t->tv_sec = seconds;
1371 *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1376 static void osd_inode_getattr(const struct lu_env *env,
1377 struct inode *inode, struct lu_attr *attr)
1379 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1380 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1381 LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE;
1383 attr->la_atime = LTIME_S(inode->i_atime);
1384 attr->la_mtime = LTIME_S(inode->i_mtime);
1385 attr->la_ctime = LTIME_S(inode->i_ctime);
1386 attr->la_mode = inode->i_mode;
1387 attr->la_size = i_size_read(inode);
1388 attr->la_blocks = inode->i_blocks;
1389 attr->la_uid = inode->i_uid;
1390 attr->la_gid = inode->i_gid;
1391 attr->la_flags = LDISKFS_I(inode)->i_flags;
1392 attr->la_nlink = inode->i_nlink;
1393 attr->la_rdev = inode->i_rdev;
1394 attr->la_blksize = 1 << inode->i_blkbits;
1395 attr->la_blkbits = inode->i_blkbits;
1398 static int osd_attr_get(const struct lu_env *env,
1399 struct dt_object *dt,
1400 struct lu_attr *attr,
1401 struct lustre_capa *capa)
1403 struct osd_object *obj = osd_dt_obj(dt);
1405 LASSERT(dt_object_exists(dt));
1406 LINVRNT(osd_invariant(obj));
1408 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1411 cfs_spin_lock(&obj->oo_guard);
1412 osd_inode_getattr(env, obj->oo_inode, attr);
1413 cfs_spin_unlock(&obj->oo_guard);
1417 static int osd_declare_attr_set(const struct lu_env *env,
1418 struct dt_object *dt,
1419 const struct lu_attr *attr,
1420 struct thandle *handle)
1422 struct osd_thandle *oh;
1423 struct osd_object *obj;
1424 struct osd_thread_info *info = osd_oti_get(env);
1425 struct lquota_id_info *qi = &info->oti_qi;
1431 LASSERT(dt != NULL);
1432 LASSERT(handle != NULL);
1434 obj = osd_dt_obj(dt);
1435 LASSERT(osd_invariant(obj));
1437 oh = container_of0(handle, struct osd_thandle, ot_super);
1438 LASSERT(oh->ot_handle == NULL);
1440 OSD_DECLARE_OP(oh, attr_set);
1441 oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
1443 if (attr == NULL || obj->oo_inode == NULL)
1446 bspace = obj->oo_inode->i_blocks;
1447 bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1448 bspace = toqb(bspace);
1450 /* Changing ownership is always preformed by super user, it should not
1453 * We still need to call the osd_declare_qid() to calculate the journal
1454 * credits for updating quota accounting files and to trigger quota
1455 * space adjustment once the operation is completed.*/
1456 if ((attr->la_valid & LA_UID) != 0 &&
1457 attr->la_uid != obj->oo_inode->i_uid) {
1458 qi->lqi_type = USRQUOTA;
1460 /* inode accounting */
1461 qi->lqi_is_blk = false;
1463 /* one more inode for the new owner ... */
1464 qi->lqi_id.qid_uid = attr->la_uid;
1466 allocated = (attr->la_uid == 0) ? true : false;
1467 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1468 if (rc == -EDQUOT || rc == -EINPROGRESS)
1473 /* and one less inode for the current uid */
1474 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1476 rc = osd_declare_qid(env, oh, qi, true, NULL);
1477 if (rc == -EDQUOT || rc == -EINPROGRESS)
1482 /* block accounting */
1483 qi->lqi_is_blk = true;
1485 /* more blocks for the new owner ... */
1486 qi->lqi_id.qid_uid = attr->la_uid;
1487 qi->lqi_space = bspace;
1488 allocated = (attr->la_uid == 0) ? true : false;
1489 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1490 if (rc == -EDQUOT || rc == -EINPROGRESS)
1495 /* and finally less blocks for the current owner */
1496 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1497 qi->lqi_space = -bspace;
1498 rc = osd_declare_qid(env, oh, qi, true, NULL);
1499 if (rc == -EDQUOT || rc == -EINPROGRESS)
1505 if (attr->la_valid & LA_GID &&
1506 attr->la_gid != obj->oo_inode->i_gid) {
1507 qi->lqi_type = GRPQUOTA;
1509 /* inode accounting */
1510 qi->lqi_is_blk = false;
1512 /* one more inode for the new group owner ... */
1513 qi->lqi_id.qid_gid = attr->la_gid;
1515 allocated = (attr->la_gid == 0) ? true : false;
1516 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1517 if (rc == -EDQUOT || rc == -EINPROGRESS)
1522 /* and one less inode for the current gid */
1523 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1525 rc = osd_declare_qid(env, oh, qi, true, NULL);
1526 if (rc == -EDQUOT || rc == -EINPROGRESS)
1531 /* block accounting */
1532 qi->lqi_is_blk = true;
1534 /* more blocks for the new owner ... */
1535 qi->lqi_id.qid_gid = attr->la_gid;
1536 qi->lqi_space = bspace;
1537 allocated = (attr->la_gid == 0) ? true : false;
1538 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1539 if (rc == -EDQUOT || rc == -EINPROGRESS)
1544 /* and finally less blocks for the current owner */
1545 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1546 qi->lqi_space = -bspace;
1547 rc = osd_declare_qid(env, oh, qi, true, NULL);
1548 if (rc == -EDQUOT || rc == -EINPROGRESS)
1557 static int osd_inode_setattr(const struct lu_env *env,
1558 struct inode *inode, const struct lu_attr *attr)
1562 bits = attr->la_valid;
1564 LASSERT(!(bits & LA_TYPE)); /* Huh? You want too much. */
1566 if (bits & LA_ATIME)
1567 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
1568 if (bits & LA_CTIME)
1569 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
1570 if (bits & LA_MTIME)
1571 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
1572 if (bits & LA_SIZE) {
1573 LDISKFS_I(inode)->i_disksize = attr->la_size;
1574 i_size_write(inode, attr->la_size);
1578 /* OSD should not change "i_blocks" which is used by quota.
1579 * "i_blocks" should be changed by ldiskfs only. */
1580 if (bits & LA_BLOCKS)
1581 inode->i_blocks = attr->la_blocks;
1584 inode->i_mode = (inode->i_mode & S_IFMT) |
1585 (attr->la_mode & ~S_IFMT);
1587 inode->i_uid = attr->la_uid;
1589 inode->i_gid = attr->la_gid;
1590 if (bits & LA_NLINK)
1591 inode->i_nlink = attr->la_nlink;
1593 inode->i_rdev = attr->la_rdev;
1595 if (bits & LA_FLAGS) {
1596 /* always keep S_NOCMTIME */
1597 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1603 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1605 if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1606 (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1611 if (attr->la_valid & LA_UID)
1612 iattr.ia_valid |= ATTR_UID;
1613 if (attr->la_valid & LA_GID)
1614 iattr.ia_valid |= ATTR_GID;
1615 iattr.ia_uid = attr->la_uid;
1616 iattr.ia_gid = attr->la_gid;
1618 rc = ll_vfs_dq_transfer(inode, &iattr);
1620 CERROR("%s: quota transfer failed: rc = %d. Is quota "
1621 "enforcement enabled on the ldiskfs filesystem?",
1622 inode->i_sb->s_id, rc);
1629 static int osd_attr_set(const struct lu_env *env,
1630 struct dt_object *dt,
1631 const struct lu_attr *attr,
1632 struct thandle *handle,
1633 struct lustre_capa *capa)
1635 struct osd_object *obj = osd_dt_obj(dt);
1636 struct inode *inode;
1639 LASSERT(handle != NULL);
1640 LASSERT(dt_object_exists(dt));
1641 LASSERT(osd_invariant(obj));
1643 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1646 OSD_EXEC_OP(handle, attr_set);
1648 inode = obj->oo_inode;
1649 if (!osd_dt_dev(handle->th_dev)->od_is_md) {
1651 rc = osd_quota_transfer(inode, attr);
1655 #ifdef HAVE_QUOTA_SUPPORT
1656 if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1657 (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1658 struct osd_ctxt *save = &osd_oti_get(env)->oti_ctxt;
1663 if (attr->la_valid & LA_UID)
1664 iattr.ia_valid |= ATTR_UID;
1665 if (attr->la_valid & LA_GID)
1666 iattr.ia_valid |= ATTR_GID;
1667 iattr.ia_uid = attr->la_uid;
1668 iattr.ia_gid = attr->la_gid;
1669 osd_push_ctxt(env, save, 1);
1670 rc = ll_vfs_dq_transfer(inode, &iattr) ? -EDQUOT : 0;
1671 osd_pop_ctxt(save, 1);
1677 cfs_spin_lock(&obj->oo_guard);
1678 rc = osd_inode_setattr(env, inode, attr);
1679 cfs_spin_unlock(&obj->oo_guard);
1682 inode->i_sb->s_op->dirty_inode(inode);
1686 struct dentry *osd_child_dentry_get(const struct lu_env *env,
1687 struct osd_object *obj,
1688 const char *name, const int namelen)
1690 return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
1693 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1695 struct dt_allocation_hint *hint,
1699 struct osd_device *osd = osd_obj2dev(obj);
1700 struct osd_thandle *oth;
1701 struct dt_object *parent = NULL;
1702 struct inode *inode;
1703 #ifdef HAVE_QUOTA_SUPPORT
1704 struct osd_ctxt *save = &info->oti_ctxt;
1707 LINVRNT(osd_invariant(obj));
1708 LASSERT(obj->oo_inode == NULL);
1709 LASSERT(obj->oo_hl_head == NULL);
1711 if (S_ISDIR(mode) && ldiskfs_pdo) {
1712 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1713 if (obj->oo_hl_head == NULL)
1717 oth = container_of(th, struct osd_thandle, ot_super);
1718 LASSERT(oth->ot_handle->h_transaction != NULL);
1720 if (hint && hint->dah_parent)
1721 parent = hint->dah_parent;
1723 #ifdef HAVE_QUOTA_SUPPORT
1724 osd_push_ctxt(info->oti_env, save, osd_dt_dev(th->th_dev)->od_is_md);
1726 inode = ldiskfs_create_inode(oth->ot_handle,
1727 parent ? osd_dt_obj(parent)->oo_inode :
1728 osd_sb(osd)->s_root->d_inode,
1730 #ifdef HAVE_QUOTA_SUPPORT
1731 osd_pop_ctxt(save, osd_dt_dev(th->th_dev)->od_is_md);
1733 if (!IS_ERR(inode)) {
1734 /* Do not update file c/mtime in ldiskfs.
1735 * NB: don't need any lock because no contention at this
1737 inode->i_flags |= S_NOCMTIME;
1738 inode->i_state |= I_LUSTRE_NOSCRUB;
1739 obj->oo_inode = inode;
1742 if (obj->oo_hl_head != NULL) {
1743 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1744 obj->oo_hl_head = NULL;
1746 result = PTR_ERR(inode);
1748 LINVRNT(osd_invariant(obj));
1756 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1757 struct lu_attr *attr,
1758 struct dt_allocation_hint *hint,
1759 struct dt_object_format *dof,
1763 struct osd_thandle *oth;
1764 struct osd_device *osd = osd_obj2dev(obj);
1765 __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1767 LASSERT(S_ISDIR(attr->la_mode));
1769 oth = container_of(th, struct osd_thandle, ot_super);
1770 LASSERT(oth->ot_handle->h_transaction != NULL);
1771 result = osd_mkfile(info, obj, mode, hint, th);
1772 if (result == 0 && osd->od_iop_mode == 0) {
1773 LASSERT(obj->oo_inode != NULL);
1775 * XXX uh-oh... call low-level iam function directly.
1778 result = iam_lvar_create(obj->oo_inode, OSD_NAME_LEN, 4,
1779 sizeof (struct osd_fid_pack),
1785 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1786 struct lu_attr *attr,
1787 struct dt_allocation_hint *hint,
1788 struct dt_object_format *dof,
1792 struct osd_thandle *oth;
1793 const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1795 __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
1797 LASSERT(S_ISREG(attr->la_mode));
1799 oth = container_of(th, struct osd_thandle, ot_super);
1800 LASSERT(oth->ot_handle->h_transaction != NULL);
1802 result = osd_mkfile(info, obj, mode, hint, th);
1804 LASSERT(obj->oo_inode != NULL);
1805 if (feat->dif_flags & DT_IND_VARKEY)
1806 result = iam_lvar_create(obj->oo_inode,
1807 feat->dif_keysize_max,
1809 feat->dif_recsize_max,
1812 result = iam_lfix_create(obj->oo_inode,
1813 feat->dif_keysize_max,
1815 feat->dif_recsize_max,
1822 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1823 struct lu_attr *attr,
1824 struct dt_allocation_hint *hint,
1825 struct dt_object_format *dof,
1828 LASSERT(S_ISREG(attr->la_mode));
1829 return osd_mkfile(info, obj, (attr->la_mode &
1830 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1833 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1834 struct lu_attr *attr,
1835 struct dt_allocation_hint *hint,
1836 struct dt_object_format *dof,
1839 LASSERT(S_ISLNK(attr->la_mode));
1840 return osd_mkfile(info, obj, (attr->la_mode &
1841 (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1844 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1845 struct lu_attr *attr,
1846 struct dt_allocation_hint *hint,
1847 struct dt_object_format *dof,
1850 cfs_umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
1853 LINVRNT(osd_invariant(obj));
1854 LASSERT(obj->oo_inode == NULL);
1855 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1856 S_ISFIFO(mode) || S_ISSOCK(mode));
1858 result = osd_mkfile(info, obj, mode, hint, th);
1860 LASSERT(obj->oo_inode != NULL);
1862 * This inode should be marked dirty for i_rdev. Currently
1863 * that is done in the osd_attr_init().
1865 init_special_inode(obj->oo_inode, mode, attr->la_rdev);
1867 LINVRNT(osd_invariant(obj));
1871 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1873 struct dt_allocation_hint *hint,
1874 struct dt_object_format *dof,
1877 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1879 osd_obj_type_f result;
1895 result = osd_mk_index;
1906 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1907 struct dt_object *parent, struct dt_object *child,
1908 cfs_umode_t child_mode)
1912 memset(ah, 0, sizeof(*ah));
1913 ah->dah_parent = parent;
1914 ah->dah_mode = child_mode;
1917 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
1918 struct lu_attr *attr, struct dt_object_format *dof)
1920 struct inode *inode = obj->oo_inode;
1921 __u64 valid = attr->la_valid;
1924 attr->la_valid &= ~(LA_TYPE | LA_MODE);
1926 if (dof->dof_type != DFT_NODE)
1927 attr->la_valid &= ~LA_RDEV;
1928 if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
1929 attr->la_valid &= ~LA_ATIME;
1930 if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
1931 attr->la_valid &= ~LA_CTIME;
1932 if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
1933 attr->la_valid &= ~LA_MTIME;
1935 if (!osd_obj2dev(obj)->od_is_md) {
1937 result = osd_quota_transfer(inode, attr);
1941 #ifdef HAVE_QUOTA_SUPPORT
1942 attr->la_valid &= ~(LA_UID | LA_GID);
1946 if (attr->la_valid != 0) {
1947 result = osd_inode_setattr(info->oti_env, inode, attr);
1949 * The osd_inode_setattr() should always succeed here. The
1950 * only error that could be returned is EDQUOT when we are
1951 * trying to change the UID or GID of the inode. However, this
1952 * should not happen since quota enforcement is no longer
1953 * enabled on ldiskfs (lquota takes care of it).
1955 LASSERTF(result == 0, "%d", result);
1956 inode->i_sb->s_op->dirty_inode(inode);
1959 attr->la_valid = valid;
1963 * Helper function for osd_object_create()
1965 * \retval 0, on success
1967 static int __osd_object_create(struct osd_thread_info *info,
1968 struct osd_object *obj, struct lu_attr *attr,
1969 struct dt_allocation_hint *hint,
1970 struct dt_object_format *dof,
1976 /* we drop umask so that permissions we pass are not affected */
1977 umask = current->fs->umask;
1978 current->fs->umask = 0;
1980 result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
1983 osd_attr_init(info, obj, attr, dof);
1984 osd_object_init0(obj);
1986 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
1987 unlock_new_inode(obj->oo_inode);
1990 /* restore previous umask value */
1991 current->fs->umask = umask;
1997 * Helper function for osd_object_create()
1999 * \retval 0, on success
2001 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2002 const struct lu_fid *fid, struct thandle *th)
2004 struct osd_thread_info *info = osd_oti_get(env);
2005 struct osd_inode_id *id = &info->oti_id;
2006 struct osd_device *osd = osd_obj2dev(obj);
2008 LASSERT(obj->oo_inode != NULL);
2010 if (osd->od_is_md) {
2011 struct md_ucred *uc = md_ucred(env);
2012 LASSERT(uc != NULL);
2015 osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2016 return osd_oi_insert(info, osd, fid, id, th);
2019 static int osd_declare_object_create(const struct lu_env *env,
2020 struct dt_object *dt,
2021 struct lu_attr *attr,
2022 struct dt_allocation_hint *hint,
2023 struct dt_object_format *dof,
2024 struct thandle *handle)
2026 struct osd_thandle *oh;
2030 LASSERT(handle != NULL);
2032 oh = container_of0(handle, struct osd_thandle, ot_super);
2033 LASSERT(oh->ot_handle == NULL);
2035 OSD_DECLARE_OP(oh, create);
2036 oh->ot_credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
2037 /* XXX: So far, only normal fid needs be inserted into the oi,
2038 * things could be changed later. Revise following code then. */
2039 if (fid_is_norm(lu_object_fid(&dt->do_lu))) {
2040 OSD_DECLARE_OP(oh, insert);
2041 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
2042 /* Reuse idle OI block may cause additional one OI block
2044 oh->ot_credits += 1;
2046 /* If this is directory, then we expect . and .. to be inserted as
2047 * well. The one directory block always needs to be created for the
2048 * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2049 * block), there is no danger of needing a tree for the first block.
2051 if (attr && S_ISDIR(attr->la_mode)) {
2052 OSD_DECLARE_OP(oh, insert);
2053 OSD_DECLARE_OP(oh, insert);
2054 oh->ot_credits += osd_dto_credits_noquota[DTO_WRITE_BASE];
2060 rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2061 false, false, NULL, false);
2065 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2066 struct lu_attr *attr,
2067 struct dt_allocation_hint *hint,
2068 struct dt_object_format *dof,
2071 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2072 struct osd_object *obj = osd_dt_obj(dt);
2073 struct osd_thread_info *info = osd_oti_get(env);
2078 LINVRNT(osd_invariant(obj));
2079 LASSERT(!dt_object_exists(dt));
2080 LASSERT(osd_write_locked(env, obj));
2081 LASSERT(th != NULL);
2083 if (unlikely(fid_is_acct(fid)))
2084 /* Quota files can't be created from the kernel any more,
2085 * 'tune2fs -O quota' will take care of creating them */
2088 OSD_EXEC_OP(th, create);
2090 result = __osd_object_create(info, obj, attr, hint, dof, th);
2092 result = __osd_oi_insert(env, obj, fid, th);
2094 LASSERT(ergo(result == 0, dt_object_exists(dt)));
2095 LASSERT(osd_invariant(obj));
2100 * Called to destroy on-disk representation of the object
2102 * Concurrency: must be locked
2104 static int osd_declare_object_destroy(const struct lu_env *env,
2105 struct dt_object *dt,
2108 struct osd_object *obj = osd_dt_obj(dt);
2109 struct inode *inode = obj->oo_inode;
2110 struct osd_thandle *oh;
2114 oh = container_of0(th, struct osd_thandle, ot_super);
2115 LASSERT(oh->ot_handle == NULL);
2118 OSD_DECLARE_OP(oh, destroy);
2119 OSD_DECLARE_OP(oh, delete);
2120 oh->ot_credits += osd_dto_credits_noquota[DTO_OBJECT_DELETE];
2121 /* XXX: So far, only normal fid needs to be inserted into the OI,
2122 * so only normal fid needs to be removed from the OI also. */
2123 if (fid_is_norm(lu_object_fid(&dt->do_lu))) {
2124 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
2125 /* Recycle idle OI leaf may cause additional three OI blocks
2127 oh->ot_credits += 3;
2130 /* one less inode */
2131 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
2132 false, true, NULL, false);
2135 /* data to be truncated */
2136 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh, true,
2141 static int osd_object_destroy(const struct lu_env *env,
2142 struct dt_object *dt,
2145 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2146 struct osd_object *obj = osd_dt_obj(dt);
2147 struct inode *inode = obj->oo_inode;
2148 struct osd_device *osd = osd_obj2dev(obj);
2149 struct osd_thandle *oh;
2153 oh = container_of0(th, struct osd_thandle, ot_super);
2154 LASSERT(oh->ot_handle);
2156 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2158 if (unlikely(fid_is_acct(fid)))
2161 /* Parallel control for OI scrub. For most of cases, there is no
2162 * lock contention. So it will not affect unlink performance. */
2163 cfs_mutex_lock(&inode->i_mutex);
2164 if (S_ISDIR(inode->i_mode)) {
2165 LASSERT(osd_inode_unlinked(inode) ||
2166 inode->i_nlink == 1);
2167 cfs_spin_lock(&obj->oo_guard);
2169 cfs_spin_unlock(&obj->oo_guard);
2170 inode->i_sb->s_op->dirty_inode(inode);
2172 LASSERT(osd_inode_unlinked(inode));
2175 OSD_EXEC_OP(th, destroy);
2177 result = osd_oi_delete(osd_oti_get(env), osd, fid, th);
2178 cfs_mutex_unlock(&inode->i_mutex);
2180 /* XXX: add to ext3 orphan list */
2181 /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2183 /* not needed in the cache anymore */
2184 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2190 * Helper function for osd_xattr_set()
2192 static int __osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2193 const struct lu_buf *buf, const char *name, int fl)
2195 struct osd_object *obj = osd_dt_obj(dt);
2196 struct inode *inode = obj->oo_inode;
2197 struct osd_thread_info *info = osd_oti_get(env);
2198 struct dentry *dentry = &info->oti_child_dentry;
2202 LASSERT(dt_object_exists(dt));
2203 LASSERT(inode->i_op != NULL && inode->i_op->setxattr != NULL);
2205 if (fl & LU_XATTR_REPLACE)
2206 fs_flags |= XATTR_REPLACE;
2208 if (fl & LU_XATTR_CREATE)
2209 fs_flags |= XATTR_CREATE;
2211 dentry->d_inode = inode;
2212 rc = inode->i_op->setxattr(dentry, name, buf->lb_buf,
2213 buf->lb_len, fs_flags);
2218 * Put the fid into lustre_mdt_attrs, and then place the structure
2219 * inode's ea. This fid should not be altered during the life time
2222 * \retval +ve, on success
2223 * \retval -ve, on error
2225 * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2227 static int osd_ea_fid_set(const struct lu_env *env, struct dt_object *dt,
2228 const struct lu_fid *fid)
2230 struct osd_thread_info *info = osd_oti_get(env);
2231 struct lustre_mdt_attrs *mdt_attrs = &info->oti_mdt_attrs;
2233 lustre_lma_init(mdt_attrs, fid);
2234 lustre_lma_swab(mdt_attrs);
2235 return __osd_xattr_set(env, dt,
2236 osd_buf_get(env, mdt_attrs, sizeof *mdt_attrs),
2237 XATTR_NAME_LMA, LU_XATTR_CREATE);
2242 * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2243 * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2244 * To have compatilibility with 1.8 ldiskfs driver we need to have
2245 * magic number at start of fid data.
2246 * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2249 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2250 const struct dt_rec *fid)
2252 param->edp_magic = LDISKFS_LUFID_MAGIC;
2253 param->edp_len = sizeof(struct lu_fid) + 1;
2255 fid_cpu_to_be((struct lu_fid *)param->edp_data,
2256 (struct lu_fid *)fid);
2260 * Try to read the fid from inode ea into dt_rec, if return value
2261 * i.e. rc is +ve, then we got fid, otherwise we will have to form igif
2263 * \param fid object fid.
2265 * \retval 0 on success
2267 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2268 __u32 ino, struct lu_fid *fid,
2269 struct osd_inode_id *id)
2271 struct osd_thread_info *info = osd_oti_get(env);
2272 struct inode *inode;
2275 osd_id_gen(id, ino, OSD_OII_NOGEN);
2276 inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2278 RETURN(PTR_ERR(inode));
2285 * OSD layer object create function for interoperability mode (b11826).
2286 * This is mostly similar to osd_object_create(). Only difference being, fid is
2287 * inserted into inode ea here.
2289 * \retval 0, on success
2290 * \retval -ve, on error
2292 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2293 struct lu_attr *attr,
2294 struct dt_allocation_hint *hint,
2295 struct dt_object_format *dof,
2298 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2299 struct osd_object *obj = osd_dt_obj(dt);
2300 struct osd_thread_info *info = osd_oti_get(env);
2305 LASSERT(osd_invariant(obj));
2306 LASSERT(!dt_object_exists(dt));
2307 LASSERT(osd_write_locked(env, obj));
2308 LASSERT(th != NULL);
2310 if (unlikely(fid_is_acct(fid)))
2311 /* Quota files can't be created from the kernel any more,
2312 * 'tune2fs -O quota' will take care of creating them */
2315 OSD_EXEC_OP(th, create);
2317 result = __osd_object_create(info, obj, attr, hint, dof, th);
2318 /* objects under osd root shld have igif fid, so dont add fid EA */
2319 if (result == 0 && fid_seq(fid) >= FID_SEQ_NORMAL)
2320 result = osd_ea_fid_set(env, dt, fid);
2323 result = __osd_oi_insert(env, obj, fid, th);
2325 LASSERT(ergo(result == 0, dt_object_exists(dt)));
2326 LINVRNT(osd_invariant(obj));
2330 static int osd_declare_object_ref_add(const struct lu_env *env,
2331 struct dt_object *dt,
2332 struct thandle *handle)
2334 struct osd_thandle *oh;
2336 /* it's possible that object doesn't exist yet */
2337 LASSERT(handle != NULL);
2339 oh = container_of0(handle, struct osd_thandle, ot_super);
2340 LASSERT(oh->ot_handle == NULL);
2342 OSD_DECLARE_OP(oh, ref_add);
2343 oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
2349 * Concurrency: @dt is write locked.
2351 static int osd_object_ref_add(const struct lu_env *env,
2352 struct dt_object *dt, struct thandle *th)
2354 struct osd_object *obj = osd_dt_obj(dt);
2355 struct inode *inode = obj->oo_inode;
2357 LINVRNT(osd_invariant(obj));
2358 LASSERT(dt_object_exists(dt));
2359 LASSERT(osd_write_locked(env, obj));
2360 LASSERT(th != NULL);
2362 OSD_EXEC_OP(th, ref_add);
2365 * DIR_NLINK feature is set for compatibility reasons if:
2366 * 1) nlinks > LDISKFS_LINK_MAX, or
2367 * 2) nlinks == 2, since this indicates i_nlink was previously 1.
2369 * It is easier to always set this flag (rather than check and set),
2370 * since it has less overhead, and the superblock will be dirtied
2371 * at some point. Both e2fsprogs and any Lustre-supported ldiskfs
2372 * do not actually care whether this flag is set or not.
2374 cfs_spin_lock(&obj->oo_guard);
2376 if (S_ISDIR(inode->i_mode) && inode->i_nlink > 1) {
2377 if (inode->i_nlink >= LDISKFS_LINK_MAX ||
2378 inode->i_nlink == 2)
2381 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2382 cfs_spin_unlock(&obj->oo_guard);
2383 inode->i_sb->s_op->dirty_inode(inode);
2384 LINVRNT(osd_invariant(obj));
2389 static int osd_declare_object_ref_del(const struct lu_env *env,
2390 struct dt_object *dt,
2391 struct thandle *handle)
2393 struct osd_thandle *oh;
2395 LASSERT(dt_object_exists(dt));
2396 LASSERT(handle != NULL);
2398 oh = container_of0(handle, struct osd_thandle, ot_super);
2399 LASSERT(oh->ot_handle == NULL);
2401 OSD_DECLARE_OP(oh, ref_del);
2402 oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
2408 * Concurrency: @dt is write locked.
2410 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2413 struct osd_object *obj = osd_dt_obj(dt);
2414 struct inode *inode = obj->oo_inode;
2416 LINVRNT(osd_invariant(obj));
2417 LASSERT(dt_object_exists(dt));
2418 LASSERT(osd_write_locked(env, obj));
2419 LASSERT(th != NULL);
2421 OSD_EXEC_OP(th, ref_del);
2423 cfs_spin_lock(&obj->oo_guard);
2424 LASSERT(inode->i_nlink > 0);
2426 /* If this is/was a many-subdir directory (nlink > LDISKFS_LINK_MAX)
2427 * then the nlink count is 1. Don't let it be set to 0 or the directory
2428 * inode will be deleted incorrectly. */
2429 if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
2431 cfs_spin_unlock(&obj->oo_guard);
2432 inode->i_sb->s_op->dirty_inode(inode);
2433 LINVRNT(osd_invariant(obj));
2439 * Get the 64-bit version for an inode.
2441 static int osd_object_version_get(const struct lu_env *env,
2442 struct dt_object *dt, dt_obj_version_t *ver)
2444 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2446 CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2447 LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2448 *ver = LDISKFS_I(inode)->i_fs_version;
2453 * Concurrency: @dt is read locked.
2455 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2456 struct lu_buf *buf, const char *name,
2457 struct lustre_capa *capa)
2459 struct osd_object *obj = osd_dt_obj(dt);
2460 struct inode *inode = obj->oo_inode;
2461 struct osd_thread_info *info = osd_oti_get(env);
2462 struct dentry *dentry = &info->oti_obj_dentry;
2464 /* version get is not real XATTR but uses xattr API */
2465 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2466 /* for version we are just using xattr API but change inode
2468 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2469 osd_object_version_get(env, dt, buf->lb_buf);
2470 return sizeof(dt_obj_version_t);
2473 LASSERT(dt_object_exists(dt));
2474 LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2476 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2479 dentry->d_inode = inode;
2480 return inode->i_op->getxattr(dentry, name, buf->lb_buf, buf->lb_len);
2484 static int osd_declare_xattr_set(const struct lu_env *env,
2485 struct dt_object *dt,
2486 const struct lu_buf *buf, const char *name,
2487 int fl, struct thandle *handle)
2489 struct osd_thandle *oh;
2491 LASSERT(handle != NULL);
2493 oh = container_of0(handle, struct osd_thandle, ot_super);
2494 LASSERT(oh->ot_handle == NULL);
2496 OSD_DECLARE_OP(oh, xattr_set);
2497 if (strcmp(name, XATTR_NAME_VERSION) == 0)
2498 oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
2500 oh->ot_credits += osd_dto_credits_noquota[DTO_XATTR_SET];
2506 * Set the 64-bit version for object
2508 static void osd_object_version_set(const struct lu_env *env,
2509 struct dt_object *dt,
2510 dt_obj_version_t *new_version)
2512 struct inode *inode = osd_dt_obj(dt)->oo_inode;
2514 CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2515 *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2517 LDISKFS_I(inode)->i_fs_version = *new_version;
2518 /** Version is set after all inode operations are finished,
2519 * so we should mark it dirty here */
2520 inode->i_sb->s_op->dirty_inode(inode);
2524 * Concurrency: @dt is write locked.
2526 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2527 const struct lu_buf *buf, const char *name, int fl,
2528 struct thandle *handle, struct lustre_capa *capa)
2530 LASSERT(handle != NULL);
2532 /* version set is not real XATTR */
2533 if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2534 /* for version we are just using xattr API but change inode
2536 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2537 osd_object_version_set(env, dt, buf->lb_buf);
2538 return sizeof(dt_obj_version_t);
2541 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2544 OSD_EXEC_OP(handle, xattr_set);
2545 return __osd_xattr_set(env, dt, buf, name, fl);
2549 * Concurrency: @dt is read locked.
2551 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
2552 struct lu_buf *buf, struct lustre_capa *capa)
2554 struct osd_object *obj = osd_dt_obj(dt);
2555 struct inode *inode = obj->oo_inode;
2556 struct osd_thread_info *info = osd_oti_get(env);
2557 struct dentry *dentry = &info->oti_obj_dentry;
2559 LASSERT(dt_object_exists(dt));
2560 LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
2561 LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2563 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2566 dentry->d_inode = inode;
2567 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2570 static int osd_declare_xattr_del(const struct lu_env *env,
2571 struct dt_object *dt, const char *name,
2572 struct thandle *handle)
2574 struct osd_thandle *oh;
2576 LASSERT(dt_object_exists(dt));
2577 LASSERT(handle != NULL);
2579 oh = container_of0(handle, struct osd_thandle, ot_super);
2580 LASSERT(oh->ot_handle == NULL);
2582 OSD_DECLARE_OP(oh, xattr_set);
2583 oh->ot_credits += osd_dto_credits_noquota[DTO_XATTR_SET];
2589 * Concurrency: @dt is write locked.
2591 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
2592 const char *name, struct thandle *handle,
2593 struct lustre_capa *capa)
2595 struct osd_object *obj = osd_dt_obj(dt);
2596 struct inode *inode = obj->oo_inode;
2597 struct osd_thread_info *info = osd_oti_get(env);
2598 struct dentry *dentry = &info->oti_obj_dentry;
2601 LASSERT(dt_object_exists(dt));
2602 LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
2603 LASSERT(osd_write_locked(env, obj));
2604 LASSERT(handle != NULL);
2606 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2609 OSD_EXEC_OP(handle, xattr_set);
2611 dentry->d_inode = inode;
2612 rc = inode->i_op->removexattr(dentry, name);
2616 static struct obd_capa *osd_capa_get(const struct lu_env *env,
2617 struct dt_object *dt,
2618 struct lustre_capa *old,
2621 struct osd_thread_info *info = osd_oti_get(env);
2622 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2623 struct osd_object *obj = osd_dt_obj(dt);
2624 struct osd_device *dev = osd_obj2dev(obj);
2625 struct lustre_capa_key *key = &info->oti_capa_key;
2626 struct lustre_capa *capa = &info->oti_capa;
2627 struct obd_capa *oc;
2628 struct md_capainfo *ci;
2632 if (!dev->od_fl_capa)
2633 RETURN(ERR_PTR(-ENOENT));
2635 LASSERT(dt_object_exists(dt));
2636 LINVRNT(osd_invariant(obj));
2638 /* renewal sanity check */
2639 if (old && osd_object_auth(env, dt, old, opc))
2640 RETURN(ERR_PTR(-EACCES));
2642 ci = md_capainfo(env);
2644 RETURN(ERR_PTR(-ENOENT));
2646 switch (ci->mc_auth) {
2650 capa->lc_uid = obj->oo_inode->i_uid;
2651 capa->lc_gid = obj->oo_inode->i_gid;
2652 capa->lc_flags = LC_ID_PLAIN;
2654 case LC_ID_CONVERT: {
2657 s[0] = obj->oo_inode->i_uid;
2658 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
2659 s[2] = obj->oo_inode->i_gid;
2660 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
2661 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
2663 RETURN(ERR_PTR(rc));
2665 capa->lc_uid = ((__u64)d[1] << 32) | d[0];
2666 capa->lc_gid = ((__u64)d[3] << 32) | d[2];
2667 capa->lc_flags = LC_ID_CONVERT;
2671 RETURN(ERR_PTR(-EINVAL));
2674 capa->lc_fid = *fid;
2676 capa->lc_flags |= dev->od_capa_alg << 24;
2677 capa->lc_timeout = dev->od_capa_timeout;
2678 capa->lc_expiry = 0;
2680 oc = capa_lookup(dev->od_capa_hash, capa, 1);
2682 LASSERT(!capa_is_expired(oc));
2686 cfs_spin_lock(&capa_lock);
2687 *key = dev->od_capa_keys[1];
2688 cfs_spin_unlock(&capa_lock);
2690 capa->lc_keyid = key->lk_keyid;
2691 capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
2693 rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
2695 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
2696 RETURN(ERR_PTR(rc));
2699 oc = capa_add(dev->od_capa_hash, capa);
2703 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
2705 struct osd_object *obj = osd_dt_obj(dt);
2706 struct inode *inode = obj->oo_inode;
2707 struct osd_thread_info *info = osd_oti_get(env);
2708 struct dentry *dentry = &info->oti_obj_dentry;
2709 struct file *file = &info->oti_file;
2714 dentry->d_inode = inode;
2715 file->f_dentry = dentry;
2716 file->f_mapping = inode->i_mapping;
2717 file->f_op = inode->i_fop;
2718 mutex_lock(&inode->i_mutex);
2719 rc = file->f_op->fsync(file, dentry, 0);
2720 mutex_unlock(&inode->i_mutex);
2724 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
2727 struct osd_object *obj = osd_dt_obj(dt);
2730 *data = (void *)obj->oo_inode;
2738 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
2739 const struct dt_index_features *feat)
2741 struct iam_descr *descr;
2743 if (osd_object_is_root(o))
2744 return feat == &dt_directory_features;
2746 LASSERT(o->oo_dir != NULL);
2748 descr = o->oo_dir->od_container.ic_descr;
2749 if (feat == &dt_directory_features) {
2750 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
2756 feat->dif_keysize_min <= descr->id_key_size &&
2757 descr->id_key_size <= feat->dif_keysize_max &&
2758 feat->dif_recsize_min <= descr->id_rec_size &&
2759 descr->id_rec_size <= feat->dif_recsize_max &&
2760 !(feat->dif_flags & (DT_IND_VARKEY |
2761 DT_IND_VARREC | DT_IND_NONUNQ)) &&
2762 ergo(feat->dif_flags & DT_IND_UPDATE,
2763 1 /* XXX check that object (and file system) is
2768 static int osd_iam_container_init(const struct lu_env *env,
2769 struct osd_object *obj,
2770 struct osd_directory *dir)
2772 struct iam_container *bag = &dir->od_container;
2775 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
2779 result = iam_container_setup(bag);
2781 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
2783 iam_container_fini(bag);
2790 * Concurrency: no external locking is necessary.
2792 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
2793 const struct dt_index_features *feat)
2797 struct osd_object *obj = osd_dt_obj(dt);
2798 struct osd_device *osd = osd_obj2dev(obj);
2800 LINVRNT(osd_invariant(obj));
2801 LASSERT(dt_object_exists(dt));
2803 if (osd_object_is_root(obj)) {
2804 dt->do_index_ops = &osd_index_ea_ops;
2806 } else if (feat == &dt_directory_features && osd->od_iop_mode) {
2807 dt->do_index_ops = &osd_index_ea_ops;
2808 if (S_ISDIR(obj->oo_inode->i_mode))
2813 } else if (unlikely(feat == &dt_otable_features)) {
2814 dt->do_index_ops = &osd_otable_ops;
2816 } else if (feat == &dt_acct_features) {
2817 dt->do_index_ops = &osd_acct_index_ops;
2820 } else if (!osd_has_index(obj)) {
2821 struct osd_directory *dir;
2826 cfs_spin_lock(&obj->oo_guard);
2827 if (obj->oo_dir == NULL)
2831 * Concurrent thread allocated container data.
2834 cfs_spin_unlock(&obj->oo_guard);
2836 * Now, that we have container data, serialize its
2839 cfs_down_write(&obj->oo_ext_idx_sem);
2841 * recheck under lock.
2843 if (!osd_has_index(obj))
2844 result = osd_iam_container_init(env, obj, dir);
2847 cfs_up_write(&obj->oo_ext_idx_sem);
2855 if (result == 0 && skip_iam == 0) {
2856 if (!osd_iam_index_probe(env, obj, feat))
2859 LINVRNT(osd_invariant(obj));
2864 static int osd_otable_it_attr_get(const struct lu_env *env,
2865 struct dt_object *dt,
2866 struct lu_attr *attr,
2867 struct lustre_capa *capa)
2873 static const struct dt_object_operations osd_obj_ops = {
2874 .do_read_lock = osd_object_read_lock,
2875 .do_write_lock = osd_object_write_lock,
2876 .do_read_unlock = osd_object_read_unlock,
2877 .do_write_unlock = osd_object_write_unlock,
2878 .do_write_locked = osd_object_write_locked,
2879 .do_attr_get = osd_attr_get,
2880 .do_declare_attr_set = osd_declare_attr_set,
2881 .do_attr_set = osd_attr_set,
2882 .do_ah_init = osd_ah_init,
2883 .do_declare_create = osd_declare_object_create,
2884 .do_create = osd_object_create,
2885 .do_declare_destroy = osd_declare_object_destroy,
2886 .do_destroy = osd_object_destroy,
2887 .do_index_try = osd_index_try,
2888 .do_declare_ref_add = osd_declare_object_ref_add,
2889 .do_ref_add = osd_object_ref_add,
2890 .do_declare_ref_del = osd_declare_object_ref_del,
2891 .do_ref_del = osd_object_ref_del,
2892 .do_xattr_get = osd_xattr_get,
2893 .do_declare_xattr_set = osd_declare_xattr_set,
2894 .do_xattr_set = osd_xattr_set,
2895 .do_declare_xattr_del = osd_declare_xattr_del,
2896 .do_xattr_del = osd_xattr_del,
2897 .do_xattr_list = osd_xattr_list,
2898 .do_capa_get = osd_capa_get,
2899 .do_object_sync = osd_object_sync,
2900 .do_data_get = osd_data_get,
2904 * dt_object_operations for interoperability mode
2905 * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
2907 static const struct dt_object_operations osd_obj_ea_ops = {
2908 .do_read_lock = osd_object_read_lock,
2909 .do_write_lock = osd_object_write_lock,
2910 .do_read_unlock = osd_object_read_unlock,
2911 .do_write_unlock = osd_object_write_unlock,
2912 .do_write_locked = osd_object_write_locked,
2913 .do_attr_get = osd_attr_get,
2914 .do_declare_attr_set = osd_declare_attr_set,
2915 .do_attr_set = osd_attr_set,
2916 .do_ah_init = osd_ah_init,
2917 .do_declare_create = osd_declare_object_create,
2918 .do_create = osd_object_ea_create,
2919 .do_declare_destroy = osd_declare_object_destroy,
2920 .do_destroy = osd_object_destroy,
2921 .do_index_try = osd_index_try,
2922 .do_declare_ref_add = osd_declare_object_ref_add,
2923 .do_ref_add = osd_object_ref_add,
2924 .do_declare_ref_del = osd_declare_object_ref_del,
2925 .do_ref_del = osd_object_ref_del,
2926 .do_xattr_get = osd_xattr_get,
2927 .do_declare_xattr_set = osd_declare_xattr_set,
2928 .do_xattr_set = osd_xattr_set,
2929 .do_declare_xattr_del = osd_declare_xattr_del,
2930 .do_xattr_del = osd_xattr_del,
2931 .do_xattr_list = osd_xattr_list,
2932 .do_capa_get = osd_capa_get,
2933 .do_object_sync = osd_object_sync,
2934 .do_data_get = osd_data_get,
2937 static const struct dt_object_operations osd_obj_otable_it_ops = {
2938 .do_attr_get = osd_otable_it_attr_get,
2939 .do_index_try = osd_index_try,
2942 static int osd_index_declare_iam_delete(const struct lu_env *env,
2943 struct dt_object *dt,
2944 const struct dt_key *key,
2945 struct thandle *handle)
2947 struct osd_thandle *oh;
2949 oh = container_of0(handle, struct osd_thandle, ot_super);
2950 LASSERT(oh->ot_handle == NULL);
2952 OSD_DECLARE_OP(oh, delete);
2953 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
2959 * delete a (key, value) pair from index \a dt specified by \a key
2961 * \param dt osd index object
2962 * \param key key for index
2963 * \param rec record reference
2964 * \param handle transaction handler
2967 * \retval -ve failure
2970 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
2971 const struct dt_key *key,
2972 struct thandle *handle,
2973 struct lustre_capa *capa)
2975 struct osd_thread_info *oti = osd_oti_get(env);
2976 struct osd_object *obj = osd_dt_obj(dt);
2977 struct osd_thandle *oh;
2978 struct iam_path_descr *ipd;
2979 struct iam_container *bag = &obj->oo_dir->od_container;
2984 LINVRNT(osd_invariant(obj));
2985 LASSERT(dt_object_exists(dt));
2986 LASSERT(bag->ic_object == obj->oo_inode);
2987 LASSERT(handle != NULL);
2989 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
2992 OSD_EXEC_OP(handle, delete);
2994 ipd = osd_idx_ipd_get(env, bag);
2995 if (unlikely(ipd == NULL))
2998 oh = container_of0(handle, struct osd_thandle, ot_super);
2999 LASSERT(oh->ot_handle != NULL);
3000 LASSERT(oh->ot_handle->h_transaction != NULL);
3002 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3003 /* swab quota uid/gid provided by caller */
3004 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3005 key = (const struct dt_key *)&oti->oti_quota_id;
3008 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3009 osd_ipd_put(env, bag, ipd);
3010 LINVRNT(osd_invariant(obj));
3014 static int osd_index_declare_ea_delete(const struct lu_env *env,
3015 struct dt_object *dt,
3016 const struct dt_key *key,
3017 struct thandle *handle)
3019 struct osd_thandle *oh;
3020 struct inode *inode;
3024 LASSERT(dt_object_exists(dt));
3025 LASSERT(handle != NULL);
3027 oh = container_of0(handle, struct osd_thandle, ot_super);
3028 LASSERT(oh->ot_handle == NULL);
3030 OSD_DECLARE_OP(oh, delete);
3031 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3033 inode = osd_dt_obj(dt)->oo_inode;
3036 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
3037 true, true, NULL, false);
3041 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3044 struct osd_fid_pack *rec;
3047 if (de->file_type & LDISKFS_DIRENT_LUFID) {
3048 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3049 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3055 * Index delete function for interoperability mode (b11826).
3056 * It will remove the directory entry added by osd_index_ea_insert().
3057 * This entry is needed to maintain name->fid mapping.
3059 * \param key, key i.e. file entry to be deleted
3061 * \retval 0, on success
3062 * \retval -ve, on error
3064 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3065 const struct dt_key *key,
3066 struct thandle *handle,
3067 struct lustre_capa *capa)
3069 struct osd_object *obj = osd_dt_obj(dt);
3070 struct inode *dir = obj->oo_inode;
3071 struct dentry *dentry;
3072 struct osd_thandle *oh;
3073 struct ldiskfs_dir_entry_2 *de;
3074 struct buffer_head *bh;
3075 struct htree_lock *hlock = NULL;
3080 LINVRNT(osd_invariant(obj));
3081 LASSERT(dt_object_exists(dt));
3082 LASSERT(handle != NULL);
3084 OSD_EXEC_OP(handle, delete);
3086 oh = container_of(handle, struct osd_thandle, ot_super);
3087 LASSERT(oh->ot_handle != NULL);
3088 LASSERT(oh->ot_handle->h_transaction != NULL);
3090 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3093 dentry = osd_child_dentry_get(env, obj,
3094 (char *)key, strlen((char *)key));
3096 if (obj->oo_hl_head != NULL) {
3097 hlock = osd_oti_get(env)->oti_hlock;
3098 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3099 dir, LDISKFS_HLOCK_DEL);
3101 cfs_down_write(&obj->oo_ext_idx_sem);
3104 bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3106 rc = ldiskfs_delete_entry(oh->ot_handle,
3113 ldiskfs_htree_unlock(hlock);
3115 cfs_up_write(&obj->oo_ext_idx_sem);
3117 LASSERT(osd_invariant(obj));
3122 * Lookup index for \a key and copy record to \a rec.
3124 * \param dt osd index object
3125 * \param key key for index
3126 * \param rec record reference
3128 * \retval +ve success : exact mach
3129 * \retval 0 return record with key not greater than \a key
3130 * \retval -ve failure
3132 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3133 struct dt_rec *rec, const struct dt_key *key,
3134 struct lustre_capa *capa)
3136 struct osd_object *obj = osd_dt_obj(dt);
3137 struct iam_path_descr *ipd;
3138 struct iam_container *bag = &obj->oo_dir->od_container;
3139 struct osd_thread_info *oti = osd_oti_get(env);
3140 struct iam_iterator *it = &oti->oti_idx_it;
3141 struct iam_rec *iam_rec;
3146 LASSERT(osd_invariant(obj));
3147 LASSERT(dt_object_exists(dt));
3148 LASSERT(bag->ic_object == obj->oo_inode);
3150 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3153 ipd = osd_idx_ipd_get(env, bag);
3157 /* got ipd now we can start iterator. */
3158 iam_it_init(it, bag, 0, ipd);
3160 if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3161 /* swab quota uid/gid provided by caller */
3162 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3163 key = (const struct dt_key *)&oti->oti_quota_id;
3166 rc = iam_it_get(it, (struct iam_key *)key);
3168 if (S_ISDIR(obj->oo_inode->i_mode))
3169 iam_rec = (struct iam_rec *)oti->oti_ldp;
3171 iam_rec = (struct iam_rec *) rec;
3173 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3175 if (S_ISDIR(obj->oo_inode->i_mode))
3176 osd_fid_unpack((struct lu_fid *) rec,
3177 (struct osd_fid_pack *)iam_rec);
3178 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3179 osd_quota_unpack(obj, rec);
3184 osd_ipd_put(env, bag, ipd);
3186 LINVRNT(osd_invariant(obj));
3191 static int osd_index_declare_iam_insert(const struct lu_env *env,
3192 struct dt_object *dt,
3193 const struct dt_rec *rec,
3194 const struct dt_key *key,
3195 struct thandle *handle)
3197 struct osd_thandle *oh;
3199 LASSERT(dt_object_exists(dt));
3200 LASSERT(handle != NULL);
3202 oh = container_of0(handle, struct osd_thandle, ot_super);
3203 LASSERT(oh->ot_handle == NULL);
3205 OSD_DECLARE_OP(oh, insert);
3206 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
3212 * Inserts (key, value) pair in \a dt index object.
3214 * \param dt osd index object
3215 * \param key key for index
3216 * \param rec record reference
3217 * \param th transaction handler
3220 * \retval -ve failure
3222 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3223 const struct dt_rec *rec,
3224 const struct dt_key *key, struct thandle *th,
3225 struct lustre_capa *capa, int ignore_quota)
3227 struct osd_object *obj = osd_dt_obj(dt);
3228 struct iam_path_descr *ipd;
3229 struct osd_thandle *oh;
3230 struct iam_container *bag = &obj->oo_dir->od_container;
3231 #ifdef HAVE_QUOTA_SUPPORT
3232 cfs_cap_t save = cfs_curproc_cap_pack();
3234 struct osd_thread_info *oti = osd_oti_get(env);
3235 struct iam_rec *iam_rec;
3240 LINVRNT(osd_invariant(obj));
3241 LASSERT(dt_object_exists(dt));
3242 LASSERT(bag->ic_object == obj->oo_inode);
3243 LASSERT(th != NULL);
3245 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3248 OSD_EXEC_OP(th, insert);
3250 ipd = osd_idx_ipd_get(env, bag);
3251 if (unlikely(ipd == NULL))
3254 oh = container_of0(th, struct osd_thandle, ot_super);
3255 LASSERT(oh->ot_handle != NULL);
3256 LASSERT(oh->ot_handle->h_transaction != NULL);
3257 #ifdef HAVE_QUOTA_SUPPORT
3259 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
3261 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
3263 if (S_ISDIR(obj->oo_inode->i_mode)) {
3264 iam_rec = (struct iam_rec *)oti->oti_ldp;
3265 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3266 } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3267 /* pack quota uid/gid */
3268 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3269 key = (const struct dt_key *)&oti->oti_quota_id;
3270 /* pack quota record */
3271 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3272 iam_rec = (struct iam_rec *)rec;
3274 iam_rec = (struct iam_rec *)rec;
3277 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3279 #ifdef HAVE_QUOTA_SUPPORT
3280 cfs_curproc_cap_unpack(save);
3282 osd_ipd_put(env, bag, ipd);
3283 LINVRNT(osd_invariant(obj));
3288 * Calls ldiskfs_add_entry() to add directory entry
3289 * into the directory. This is required for
3290 * interoperability mode (b11826)
3292 * \retval 0, on success
3293 * \retval -ve, on error
3295 static int __osd_ea_add_rec(struct osd_thread_info *info,
3296 struct osd_object *pobj, struct inode *cinode,
3297 const char *name, const struct dt_rec *fid,
3298 struct htree_lock *hlock, struct thandle *th)
3300 struct ldiskfs_dentry_param *ldp;
3301 struct dentry *child;
3302 struct osd_thandle *oth;
3305 oth = container_of(th, struct osd_thandle, ot_super);
3306 LASSERT(oth->ot_handle != NULL);
3307 LASSERT(oth->ot_handle->h_transaction != NULL);
3309 child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3311 /* XXX: remove fid_is_igif() check here.
3312 * IGIF check is just to handle insertion of .. when it is 'ROOT',
3313 * it is IGIF now but needs FID in dir entry as well for readdir
3315 * LU-838 should fix that and remove fid_is_igif() check */
3316 if (fid_is_igif((struct lu_fid *)fid) ||
3317 fid_is_norm((struct lu_fid *)fid)) {
3318 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3319 osd_get_ldiskfs_dirent_param(ldp, fid);
3320 child->d_fsdata = (void *)ldp;
3322 child->d_fsdata = NULL;
3324 rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3330 * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3331 * into the directory.Also sets flags into osd object to
3332 * indicate dot and dotdot are created. This is required for
3333 * interoperability mode (b11826)
3335 * \param dir directory for dot and dotdot fixup.
3336 * \param obj child object for linking
3338 * \retval 0, on success
3339 * \retval -ve, on error
3341 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3342 struct osd_object *dir,
3343 struct inode *parent_dir, const char *name,
3344 const struct dt_rec *dot_fid,
3345 const struct dt_rec *dot_dot_fid,
3348 struct inode *inode = dir->oo_inode;
3349 struct ldiskfs_dentry_param *dot_ldp;
3350 struct ldiskfs_dentry_param *dot_dot_ldp;
3351 struct osd_thandle *oth;
3354 oth = container_of(th, struct osd_thandle, ot_super);
3355 LASSERT(oth->ot_handle->h_transaction != NULL);
3356 LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3358 if (strcmp(name, dot) == 0) {
3359 if (dir->oo_compat_dot_created) {
3362 LASSERT(inode == parent_dir);
3363 dir->oo_compat_dot_created = 1;
3366 } else if(strcmp(name, dotdot) == 0) {
3367 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3368 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3370 if (!dir->oo_compat_dot_created)
3372 if (!fid_is_igif((struct lu_fid *)dot_fid)) {
3373 osd_get_ldiskfs_dirent_param(dot_ldp, dot_fid);
3374 osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3379 /* in case of rename, dotdot is already created */
3380 if (dir->oo_compat_dotdot_created) {
3381 return __osd_ea_add_rec(info, dir, parent_dir, name,
3382 dot_dot_fid, NULL, th);
3385 result = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3386 inode, dot_ldp, dot_dot_ldp);
3388 dir->oo_compat_dotdot_created = 1;
3396 * It will call the appropriate osd_add* function and return the
3397 * value, return by respective functions.
3399 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3400 struct inode *cinode, const char *name,
3401 const struct dt_rec *fid, struct thandle *th)
3403 struct osd_thread_info *info = osd_oti_get(env);
3404 struct htree_lock *hlock;
3407 hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3409 if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3411 if (hlock != NULL) {
3412 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3415 cfs_down_write(&pobj->oo_ext_idx_sem);
3417 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3418 (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
3421 if (hlock != NULL) {
3422 ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3423 pobj->oo_inode, LDISKFS_HLOCK_ADD);
3425 cfs_down_write(&pobj->oo_ext_idx_sem);
3428 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
3432 ldiskfs_htree_unlock(hlock);
3434 cfs_up_write(&pobj->oo_ext_idx_sem);
3440 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
3441 struct osd_idmap_cache *oic)
3443 struct osd_scrub *scrub = &dev->od_scrub;
3444 struct lu_fid *fid = &oic->oic_fid;
3445 struct osd_inode_id *id = &oti->oti_id;
3451 rc = osd_oi_lookup(oti, dev, fid, id);
3452 if (rc != 0 && rc != -ENOENT)
3455 if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
3458 if (thread_is_running(&scrub->os_thread)) {
3459 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
3460 /* There is race condition between osd_oi_lookup and OI scrub.
3461 * The OI scrub finished just after osd_oi_lookup() failure.
3462 * Under such case, it is unnecessary to trigger OI scrub again,
3463 * but try to call osd_oi_lookup() again. */
3464 if (unlikely(rc == -EAGAIN))
3470 if (!scrub->os_no_scrub && ++once == 1) {
3471 CDEBUG(D_LFSCK, "Trigger OI scrub by RPC for "DFID"\n",
3473 rc = osd_scrub_start(dev);
3474 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC for "DFID
3476 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
3482 RETURN(rc = -EREMCHG);
3486 * Calls ->lookup() to find dentry. From dentry get inode and
3487 * read inode's ea to get fid. This is required for interoperability
3490 * \retval 0, on success
3491 * \retval -ve, on error
3493 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
3494 struct dt_rec *rec, const struct dt_key *key)
3496 struct inode *dir = obj->oo_inode;
3497 struct dentry *dentry;
3498 struct ldiskfs_dir_entry_2 *de;
3499 struct buffer_head *bh;
3500 struct lu_fid *fid = (struct lu_fid *) rec;
3501 struct htree_lock *hlock = NULL;
3505 LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
3507 dentry = osd_child_dentry_get(env, obj,
3508 (char *)key, strlen((char *)key));
3510 if (obj->oo_hl_head != NULL) {
3511 hlock = osd_oti_get(env)->oti_hlock;
3512 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3513 dir, LDISKFS_HLOCK_LOOKUP);
3515 cfs_down_read(&obj->oo_ext_idx_sem);
3518 bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3520 struct osd_thread_info *oti = osd_oti_get(env);
3521 struct osd_idmap_cache *oic = &oti->oti_cache;
3522 struct osd_device *dev = osd_obj2dev(obj);
3523 struct osd_scrub *scrub = &dev->od_scrub;
3524 struct scrub_file *sf = &scrub->os_file;
3526 ino = le32_to_cpu(de->inode);
3527 rc = osd_get_fid_from_dentry(de, rec);
3529 /* done with de, release bh */
3532 rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid);
3534 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
3535 if (rc != 0 || !fid_is_norm(fid)) {
3536 fid_zero(&oic->oic_fid);
3540 oic->oic_fid = *fid;
3541 if ((scrub->os_pos_current <= ino) &&
3542 (sf->sf_flags & SF_INCONSISTENT ||
3543 ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
3545 rc = osd_consistency_check(oti, dev, oic);
3554 ldiskfs_htree_unlock(hlock);
3556 cfs_up_read(&obj->oo_ext_idx_sem);
3561 * Find the osd object for given fid.
3563 * \param fid need to find the osd object having this fid
3565 * \retval osd_object on success
3566 * \retval -ve on error
3568 struct osd_object *osd_object_find(const struct lu_env *env,
3569 struct dt_object *dt,
3570 const struct lu_fid *fid)
3572 struct lu_device *ludev = dt->do_lu.lo_dev;
3573 struct osd_object *child = NULL;
3574 struct lu_object *luch;
3575 struct lu_object *lo;
3578 * at this point topdev might not exist yet
3579 * (i.e. MGS is preparing profiles). so we can
3580 * not rely on topdev and instead lookup with
3581 * our device passed as topdev. this can't work
3582 * if the object isn't cached yet (as osd doesn't
3583 * allocate lu_header). IOW, the object must be
3584 * in the cache, otherwise lu_object_alloc() crashes
3587 luch = lu_object_find_at(env, ludev, fid, NULL);
3588 if (!IS_ERR(luch)) {
3589 if (lu_object_exists(luch)) {
3590 lo = lu_object_locate(luch->lo_header, ludev->ld_type);
3592 child = osd_obj(lo);
3594 LU_OBJECT_DEBUG(D_ERROR, env, luch,
3595 "lu_object can't be located"
3596 DFID"\n", PFID(fid));
3598 if (child == NULL) {
3599 lu_object_put(env, luch);
3600 CERROR("Unable to get osd_object\n");
3601 child = ERR_PTR(-ENOENT);
3604 LU_OBJECT_DEBUG(D_ERROR, env, luch,
3605 "lu_object does not exists "DFID"\n",
3607 lu_object_put(env, luch);
3608 child = ERR_PTR(-ENOENT);
3611 child = (void *)luch;
3617 * Put the osd object once done with it.
3619 * \param obj osd object that needs to be put
3621 static inline void osd_object_put(const struct lu_env *env,
3622 struct osd_object *obj)
3624 lu_object_put(env, &obj->oo_dt.do_lu);
3627 static int osd_index_declare_ea_insert(const struct lu_env *env,
3628 struct dt_object *dt,
3629 const struct dt_rec *rec,
3630 const struct dt_key *key,
3631 struct thandle *handle)
3633 struct osd_thandle *oh;
3634 struct inode *inode;
3638 LASSERT(dt_object_exists(dt));
3639 LASSERT(handle != NULL);
3641 oh = container_of0(handle, struct osd_thandle, ot_super);
3642 LASSERT(oh->ot_handle == NULL);
3644 OSD_DECLARE_OP(oh, insert);
3645 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
3647 inode = osd_dt_obj(dt)->oo_inode;
3650 /* We ignore block quota on meta pool (MDTs), so needn't
3651 * calculate how many blocks will be consumed by this index
3653 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
3654 true, true, NULL, false);
3659 * Index add function for interoperability mode (b11826).
3660 * It will add the directory entry.This entry is needed to
3661 * maintain name->fid mapping.
3663 * \param key it is key i.e. file entry to be inserted
3664 * \param rec it is value of given key i.e. fid
3666 * \retval 0, on success
3667 * \retval -ve, on error
3669 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
3670 const struct dt_rec *rec,
3671 const struct dt_key *key, struct thandle *th,
3672 struct lustre_capa *capa, int ignore_quota)
3674 struct osd_object *obj = osd_dt_obj(dt);
3675 struct lu_fid *fid = (struct lu_fid *) rec;
3676 const char *name = (const char *)key;
3677 struct osd_object *child;
3678 #ifdef HAVE_QUOTA_SUPPORT
3679 cfs_cap_t save = cfs_curproc_cap_pack();
3685 LASSERT(osd_invariant(obj));
3686 LASSERT(dt_object_exists(dt));
3687 LASSERT(th != NULL);
3689 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3692 child = osd_object_find(env, dt, fid);
3693 if (!IS_ERR(child)) {
3694 #ifdef HAVE_QUOTA_SUPPORT
3696 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
3698 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
3700 rc = osd_ea_add_rec(env, obj, child->oo_inode, name, rec, th);
3701 #ifdef HAVE_QUOTA_SUPPORT
3702 cfs_curproc_cap_unpack(save);
3704 osd_object_put(env, child);
3706 rc = PTR_ERR(child);
3709 LASSERT(osd_invariant(obj));
3714 * Initialize osd Iterator for given osd index object.
3716 * \param dt osd index object
3719 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
3720 struct dt_object *dt,
3722 struct lustre_capa *capa)
3724 struct osd_it_iam *it;
3725 struct osd_thread_info *oti = osd_oti_get(env);
3726 struct osd_object *obj = osd_dt_obj(dt);
3727 struct lu_object *lo = &dt->do_lu;
3728 struct iam_path_descr *ipd;
3729 struct iam_container *bag = &obj->oo_dir->od_container;
3731 LASSERT(lu_object_exists(lo));
3733 if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
3734 return ERR_PTR(-EACCES);
3737 ipd = osd_it_ipd_get(env, bag);
3738 if (likely(ipd != NULL)) {
3742 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
3743 return (struct dt_it *)it;
3745 return ERR_PTR(-ENOMEM);
3749 * free given Iterator.
3752 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
3754 struct osd_it_iam *it = (struct osd_it_iam *)di;
3755 struct osd_object *obj = it->oi_obj;
3757 iam_it_fini(&it->oi_it);
3758 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
3759 lu_object_put(env, &obj->oo_dt.do_lu);
3763 * Move Iterator to record specified by \a key
3765 * \param di osd iterator
3766 * \param key key for index
3768 * \retval +ve di points to record with least key not larger than key
3769 * \retval 0 di points to exact matched key
3770 * \retval -ve failure
3773 static int osd_it_iam_get(const struct lu_env *env,
3774 struct dt_it *di, const struct dt_key *key)
3776 struct osd_thread_info *oti = osd_oti_get(env);
3777 struct osd_it_iam *it = (struct osd_it_iam *)di;
3779 if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
3780 /* swab quota uid/gid */
3781 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3782 key = (struct dt_key *)&oti->oti_quota_id;
3785 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
3791 * \param di osd iterator
3794 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
3796 struct osd_it_iam *it = (struct osd_it_iam *)di;
3798 iam_it_put(&it->oi_it);
3802 * Move iterator by one record
3804 * \param di osd iterator
3806 * \retval +1 end of container reached
3808 * \retval -ve failure
3811 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
3813 struct osd_it_iam *it = (struct osd_it_iam *)di;
3815 return iam_it_next(&it->oi_it);
3819 * Return pointer to the key under iterator.
3822 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
3823 const struct dt_it *di)
3825 struct osd_thread_info *oti = osd_oti_get(env);
3826 struct osd_it_iam *it = (struct osd_it_iam *)di;
3827 struct osd_object *obj = it->oi_obj;
3830 key = (struct dt_key *)iam_it_key_get(&it->oi_it);
3832 if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
3833 /* swab quota uid/gid */
3834 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
3835 key = (struct dt_key *)&oti->oti_quota_id;
3842 * Return size of key under iterator (in bytes)
3845 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
3847 struct osd_it_iam *it = (struct osd_it_iam *)di;
3849 return iam_it_key_size(&it->oi_it);
3852 static inline void osd_it_append_attrs(struct lu_dirent *ent, __u32 attr,
3853 int len, __u16 type)
3855 struct luda_type *lt;
3856 const unsigned align = sizeof(struct luda_type) - 1;
3858 /* check if file type is required */
3859 if (attr & LUDA_TYPE) {
3860 len = (len + align) & ~align;
3862 lt = (void *) ent->lde_name + len;
3863 lt->lt_type = cpu_to_le16(CFS_DTTOIF(type));
3864 ent->lde_attrs |= LUDA_TYPE;
3867 ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
3871 * build lu direct from backend fs dirent.
3874 static inline void osd_it_pack_dirent(struct lu_dirent *ent,
3875 struct lu_fid *fid, __u64 offset,
3876 char *name, __u16 namelen,
3877 __u16 type, __u32 attr)
3879 fid_cpu_to_le(&ent->lde_fid, fid);
3880 ent->lde_attrs = LUDA_FID;
3882 ent->lde_hash = cpu_to_le64(offset);
3883 ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
3885 strncpy(ent->lde_name, name, namelen);
3886 ent->lde_namelen = cpu_to_le16(namelen);
3888 /* append lustre attributes */
3889 osd_it_append_attrs(ent, attr, namelen, type);
3893 * Return pointer to the record under iterator.
3895 static int osd_it_iam_rec(const struct lu_env *env,
3896 const struct dt_it *di,
3897 struct dt_rec *dtrec, __u32 attr)
3899 struct osd_it_iam *it = (struct osd_it_iam *)di;
3900 struct osd_thread_info *info = osd_oti_get(env);
3903 if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
3904 const struct osd_fid_pack *rec;
3905 struct lu_fid *fid = &info->oti_fid;
3906 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
3912 name = (char *)iam_it_key_get(&it->oi_it);
3914 RETURN(PTR_ERR(name));
3916 namelen = iam_it_key_size(&it->oi_it);
3918 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
3920 RETURN(PTR_ERR(rec));
3922 rc = osd_fid_unpack(fid, rec);
3926 hash = iam_it_store(&it->oi_it);
3928 /* IAM does not store object type in IAM index (dir) */
3929 osd_it_pack_dirent(lde, fid, hash, name, namelen,
3931 } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
3932 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
3933 (struct iam_rec *)dtrec);
3934 osd_quota_unpack(it->oi_obj, dtrec);
3936 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
3937 (struct iam_rec *)dtrec);
3944 * Returns cookie for current Iterator position.
3946 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
3948 struct osd_it_iam *it = (struct osd_it_iam *)di;
3950 return iam_it_store(&it->oi_it);
3954 * Restore iterator from cookie.
3956 * \param di osd iterator
3957 * \param hash Iterator location cookie
3959 * \retval +ve di points to record with least key not larger than key.
3960 * \retval 0 di points to exact matched key
3961 * \retval -ve failure
3964 static int osd_it_iam_load(const struct lu_env *env,
3965 const struct dt_it *di, __u64 hash)
3967 struct osd_it_iam *it = (struct osd_it_iam *)di;
3969 return iam_it_load(&it->oi_it, hash);
3972 static const struct dt_index_operations osd_index_iam_ops = {
3973 .dio_lookup = osd_index_iam_lookup,
3974 .dio_declare_insert = osd_index_declare_iam_insert,
3975 .dio_insert = osd_index_iam_insert,
3976 .dio_declare_delete = osd_index_declare_iam_delete,
3977 .dio_delete = osd_index_iam_delete,
3979 .init = osd_it_iam_init,
3980 .fini = osd_it_iam_fini,
3981 .get = osd_it_iam_get,
3982 .put = osd_it_iam_put,
3983 .next = osd_it_iam_next,
3984 .key = osd_it_iam_key,
3985 .key_size = osd_it_iam_key_size,
3986 .rec = osd_it_iam_rec,
3987 .store = osd_it_iam_store,
3988 .load = osd_it_iam_load
3993 * Creates or initializes iterator context.
3995 * \retval struct osd_it_ea, iterator structure on success
3998 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
3999 struct dt_object *dt,
4001 struct lustre_capa *capa)
4003 struct osd_object *obj = osd_dt_obj(dt);
4004 struct osd_thread_info *info = osd_oti_get(env);
4005 struct osd_it_ea *it = &info->oti_it_ea;
4006 struct lu_object *lo = &dt->do_lu;
4007 struct dentry *obj_dentry = &info->oti_it_dentry;
4009 LASSERT(lu_object_exists(lo));
4011 obj_dentry->d_inode = obj->oo_inode;
4012 obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4013 obj_dentry->d_name.hash = 0;
4015 it->oie_rd_dirent = 0;
4016 it->oie_it_dirent = 0;
4017 it->oie_dirent = NULL;
4018 it->oie_buf = info->oti_it_ea_buf;
4020 it->oie_file.f_pos = 0;
4021 it->oie_file.f_dentry = obj_dentry;
4022 if (attr & LUDA_64BITHASH)
4023 it->oie_file.f_mode |= FMODE_64BITHASH;
4025 it->oie_file.f_mode |= FMODE_32BITHASH;
4026 it->oie_file.f_mapping = obj->oo_inode->i_mapping;
4027 it->oie_file.f_op = obj->oo_inode->i_fop;
4028 it->oie_file.private_data = NULL;
4030 RETURN((struct dt_it *) it);
4034 * Destroy or finishes iterator context.
4036 * \param di iterator structure to be destroyed
4038 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4040 struct osd_it_ea *it = (struct osd_it_ea *)di;
4041 struct osd_object *obj = it->oie_obj;
4042 struct inode *inode = obj->oo_inode;
4045 it->oie_file.f_op->release(inode, &it->oie_file);
4046 lu_object_put(env, &obj->oo_dt.do_lu);
4051 * It position the iterator at given key, so that next lookup continues from
4052 * that key Or it is similar to dio_it->load() but based on a key,
4053 * rather than file position.
4055 * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4058 * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4060 static int osd_it_ea_get(const struct lu_env *env,
4061 struct dt_it *di, const struct dt_key *key)
4063 struct osd_it_ea *it = (struct osd_it_ea *)di;
4066 LASSERT(((const char *)key)[0] == '\0');
4067 it->oie_file.f_pos = 0;
4068 it->oie_rd_dirent = 0;
4069 it->oie_it_dirent = 0;
4070 it->oie_dirent = NULL;
4078 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4083 * It is called internally by ->readdir(). It fills the
4084 * iterator's in-memory data structure with required
4085 * information i.e. name, namelen, rec_size etc.
4087 * \param buf in which information to be filled in.
4088 * \param name name of the file in given dir
4090 * \retval 0 on success
4091 * \retval 1 on buffer full
4093 static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen,
4094 loff_t offset, __u64 ino,
4097 struct osd_it_ea *it = (struct osd_it_ea *)buf;
4098 struct osd_it_ea_dirent *ent = it->oie_dirent;
4099 struct lu_fid *fid = &ent->oied_fid;
4100 struct osd_fid_pack *rec;
4103 /* this should never happen */
4104 if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4105 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4109 if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4113 if (d_type & LDISKFS_DIRENT_LUFID) {
4114 rec = (struct osd_fid_pack*) (name + namelen + 1);
4116 if (osd_fid_unpack(fid, rec) != 0)
4119 d_type &= ~LDISKFS_DIRENT_LUFID;
4124 ent->oied_ino = ino;
4125 ent->oied_off = offset;
4126 ent->oied_namelen = namelen;
4127 ent->oied_type = d_type;
4129 memcpy(ent->oied_name, name, namelen);
4131 it->oie_rd_dirent++;
4132 it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4137 * Calls ->readdir() to load a directory entry at a time
4138 * and stored it in iterator's in-memory data structure.
4140 * \param di iterator's in memory structure
4142 * \retval 0 on success
4143 * \retval -ve on error
4145 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4146 const struct dt_it *di)
4148 struct osd_it_ea *it = (struct osd_it_ea *)di;
4149 struct osd_object *obj = it->oie_obj;
4150 struct inode *inode = obj->oo_inode;
4151 struct htree_lock *hlock = NULL;
4155 it->oie_dirent = it->oie_buf;
4156 it->oie_rd_dirent = 0;
4158 if (obj->oo_hl_head != NULL) {
4159 hlock = osd_oti_get(env)->oti_hlock;
4160 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4161 inode, LDISKFS_HLOCK_READDIR);
4163 cfs_down_read(&obj->oo_ext_idx_sem);
4166 result = inode->i_fop->readdir(&it->oie_file, it,
4167 (filldir_t) osd_ldiskfs_filldir);
4170 ldiskfs_htree_unlock(hlock);
4172 cfs_up_read(&obj->oo_ext_idx_sem);
4174 if (it->oie_rd_dirent == 0) {
4177 it->oie_dirent = it->oie_buf;
4178 it->oie_it_dirent = 1;
4185 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4186 * to load a directory entry at a time and stored it in
4187 * iterator's in-memory data structure.
4189 * \param di iterator's in memory structure
4191 * \retval +ve iterator reached to end
4192 * \retval 0 iterator not reached to end
4193 * \retval -ve on error
4195 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4197 struct osd_it_ea *it = (struct osd_it_ea *)di;
4202 if (it->oie_it_dirent < it->oie_rd_dirent) {
4204 (void *) it->oie_dirent +
4205 cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4206 it->oie_dirent->oied_namelen);
4207 it->oie_it_dirent++;
4210 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4213 rc = osd_ldiskfs_it_fill(env, di);
4220 * Returns the key at current position from iterator's in memory structure.
4222 * \param di iterator's in memory structure
4224 * \retval key i.e. struct dt_key on success
4226 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4227 const struct dt_it *di)
4229 struct osd_it_ea *it = (struct osd_it_ea *)di;
4231 return (struct dt_key *)it->oie_dirent->oied_name;
4235 * Returns the key's size at current position from iterator's in memory structure.
4237 * \param di iterator's in memory structure
4239 * \retval key_size i.e. struct dt_key on success
4241 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4243 struct osd_it_ea *it = (struct osd_it_ea *)di;
4245 return it->oie_dirent->oied_namelen;
4250 * Returns the value (i.e. fid/igif) at current position from iterator's
4251 * in memory structure.
4253 * \param di struct osd_it_ea, iterator's in memory structure
4254 * \param attr attr requested for dirent.
4255 * \param lde lustre dirent
4257 * \retval 0 no error and \param lde has correct lustre dirent.
4258 * \retval -ve on error
4260 static inline int osd_it_ea_rec(const struct lu_env *env,
4261 const struct dt_it *di,
4262 struct dt_rec *dtrec, __u32 attr)
4264 struct osd_it_ea *it = (struct osd_it_ea *)di;
4265 struct osd_object *obj = it->oie_obj;
4266 struct osd_device *dev = osd_obj2dev(obj);
4267 struct osd_scrub *scrub = &dev->od_scrub;
4268 struct scrub_file *sf = &scrub->os_file;
4269 struct osd_thread_info *oti = osd_oti_get(env);
4270 struct osd_idmap_cache *oic = &oti->oti_cache;
4271 struct lu_fid *fid = &it->oie_dirent->oied_fid;
4272 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
4273 __u32 ino = it->oie_dirent->oied_ino;
4277 if (!fid_is_sane(fid)) {
4278 rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid);
4282 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4285 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
4286 it->oie_dirent->oied_name,
4287 it->oie_dirent->oied_namelen,
4288 it->oie_dirent->oied_type, attr);
4289 if (!fid_is_norm(fid)) {
4290 fid_zero(&oic->oic_fid);
4294 oic->oic_fid = *fid;
4295 if ((scrub->os_pos_current <= ino) &&
4296 (sf->sf_flags & SF_INCONSISTENT ||
4297 ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
4298 rc = osd_consistency_check(oti, dev, oic);
4304 * Returns a cookie for current position of the iterator head, so that
4305 * user can use this cookie to load/start the iterator next time.
4307 * \param di iterator's in memory structure
4309 * \retval cookie for current position, on success
4311 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
4313 struct osd_it_ea *it = (struct osd_it_ea *)di;
4315 return it->oie_dirent->oied_off;
4319 * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4320 * to load a directory entry at a time and stored it i inn,
4321 * in iterator's in-memory data structure.
4323 * \param di struct osd_it_ea, iterator's in memory structure
4325 * \retval +ve on success
4326 * \retval -ve on error
4328 static int osd_it_ea_load(const struct lu_env *env,
4329 const struct dt_it *di, __u64 hash)
4331 struct osd_it_ea *it = (struct osd_it_ea *)di;
4335 it->oie_file.f_pos = hash;
4337 rc = osd_ldiskfs_it_fill(env, di);
4345 * Index lookup function for interoperability mode (b11826).
4347 * \param key, key i.e. file name to be searched
4349 * \retval +ve, on success
4350 * \retval -ve, on error
4352 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
4353 struct dt_rec *rec, const struct dt_key *key,
4354 struct lustre_capa *capa)
4356 struct osd_object *obj = osd_dt_obj(dt);
4361 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
4362 LINVRNT(osd_invariant(obj));
4364 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
4367 rc = osd_ea_lookup_rec(env, obj, rec, key);
4375 * Index and Iterator operations for interoperability
4376 * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
4378 static const struct dt_index_operations osd_index_ea_ops = {
4379 .dio_lookup = osd_index_ea_lookup,
4380 .dio_declare_insert = osd_index_declare_ea_insert,
4381 .dio_insert = osd_index_ea_insert,
4382 .dio_declare_delete = osd_index_declare_ea_delete,
4383 .dio_delete = osd_index_ea_delete,
4385 .init = osd_it_ea_init,
4386 .fini = osd_it_ea_fini,
4387 .get = osd_it_ea_get,
4388 .put = osd_it_ea_put,
4389 .next = osd_it_ea_next,
4390 .key = osd_it_ea_key,
4391 .key_size = osd_it_ea_key_size,
4392 .rec = osd_it_ea_rec,
4393 .store = osd_it_ea_store,
4394 .load = osd_it_ea_load
4398 static void *osd_key_init(const struct lu_context *ctx,
4399 struct lu_context_key *key)
4401 struct osd_thread_info *info;
4403 OBD_ALLOC_PTR(info);
4405 return ERR_PTR(-ENOMEM);
4407 OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4408 if (info->oti_it_ea_buf == NULL)
4411 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
4413 info->oti_hlock = ldiskfs_htree_lock_alloc();
4414 if (info->oti_hlock == NULL)
4420 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4423 return ERR_PTR(-ENOMEM);
4426 static void osd_key_fini(const struct lu_context *ctx,
4427 struct lu_context_key *key, void* data)
4429 struct osd_thread_info *info = data;
4431 if (info->oti_hlock != NULL)
4432 ldiskfs_htree_lock_free(info->oti_hlock);
4433 OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4437 static void osd_key_exit(const struct lu_context *ctx,
4438 struct lu_context_key *key, void *data)
4440 struct osd_thread_info *info = data;
4442 LASSERT(info->oti_r_locks == 0);
4443 LASSERT(info->oti_w_locks == 0);
4444 LASSERT(info->oti_txns == 0);
4447 /* type constructor/destructor: osd_type_init, osd_type_fini */
4448 LU_TYPE_INIT_FINI(osd, &osd_key);
4450 struct lu_context_key osd_key = {
4451 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
4452 .lct_init = osd_key_init,
4453 .lct_fini = osd_key_fini,
4454 .lct_exit = osd_key_exit
4458 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
4459 const char *name, struct lu_device *next)
4461 struct osd_device *osd = osd_dev(d);
4463 strncpy(osd->od_svname, name, MAX_OBD_NAME);
4464 return osd_procfs_init(osd, name);
4467 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
4471 osd_scrub_cleanup(env, o);
4474 fsfilt_put_ops(o->od_fsops);
4478 /* shutdown quota slave instance associated with the device */
4479 if (o->od_quota_slave != NULL) {
4480 qsd_fini(env, o->od_quota_slave);
4481 o->od_quota_slave = NULL;
4487 static int osd_mount(const struct lu_env *env,
4488 struct osd_device *o, struct lustre_cfg *cfg)
4490 const char *name = lustre_cfg_string(cfg, 0);
4491 const char *dev = lustre_cfg_string(cfg, 1);
4493 unsigned long page, s_flags, lmd_flags = 0;
4494 struct page *__page;
4495 struct file_system_type *type;
4496 char *options = NULL;
4501 if (o->od_mnt != NULL)
4504 o->od_fsops = fsfilt_get_ops(mt_str(LDD_MT_LDISKFS));
4505 if (o->od_fsops == NULL) {
4506 CERROR("Can't find fsfilt_ldiskfs\n");
4510 OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
4512 GOTO(out, rc = -ENOMEM);
4514 str = lustre_cfg_string(cfg, 2);
4515 s_flags = simple_strtoul(str, NULL, 0);
4516 str = strstr(str, ":");
4518 lmd_flags = simple_strtoul(str + 1, NULL, 0);
4519 opts = lustre_cfg_string(cfg, 3);
4520 page = (unsigned long)cfs_page_address(__page);
4521 options = (char *)page;
4524 strcat(options, "user_xattr,acl");
4526 strcat(options, opts);
4528 /* Glom up mount options */
4529 if (*options != '\0')
4530 strcat(options, ",");
4531 strlcat(options, "no_mbcache", CFS_PAGE_SIZE);
4533 type = get_fs_type("ldiskfs");
4535 CERROR("%s: cannot find ldiskfs module\n", name);
4536 GOTO(out, rc = -ENODEV);
4539 o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
4540 cfs_module_put(type->owner);
4542 if (IS_ERR(o->od_mnt)) {
4543 rc = PTR_ERR(o->od_mnt);
4544 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
4549 if (lvfs_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
4550 CERROR("%s: underlying device %s is marked as read-only. "
4551 "Setup failed\n", name, dev);
4554 GOTO(out, rc = -EROFS);
4557 if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
4558 LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
4559 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
4562 GOTO(out, rc = -EINVAL);
4565 if (lmd_flags & LMD_FLG_IAM) {
4567 LCONSOLE_WARN("%s: OSD: IAM mode enabled\n", name);
4570 if (lmd_flags & LMD_FLG_NOSCRUB)
4571 o->od_scrub.os_no_scrub = 1;
4575 OBD_PAGE_FREE(__page);
4577 fsfilt_put_ops(o->od_fsops);
4582 static struct lu_device *osd_device_fini(const struct lu_env *env,
4583 struct lu_device *d)
4588 rc = osd_shutdown(env, osd_dev(d));
4590 osd_compat_fini(osd_dev(d));
4592 shrink_dcache_sb(osd_sb(osd_dev(d)));
4593 osd_sync(env, lu2dt_dev(d));
4595 rc = osd_procfs_fini(osd_dev(d));
4597 CERROR("proc fini error %d \n", rc);
4598 RETURN (ERR_PTR(rc));
4601 if (osd_dev(d)->od_mnt) {
4602 mntput(osd_dev(d)->od_mnt);
4603 osd_dev(d)->od_mnt = NULL;
4609 static int osd_device_init0(const struct lu_env *env,
4610 struct osd_device *o,
4611 struct lustre_cfg *cfg)
4613 struct lu_device *l = osd2lu_dev(o);
4614 struct osd_thread_info *info;
4617 /* if the module was re-loaded, env can loose its keys */
4618 rc = lu_env_refill((struct lu_env *) env);
4621 info = osd_oti_get(env);
4624 l->ld_ops = &osd_lu_ops;
4625 o->od_dt_dev.dd_ops = &osd_dt_ops;
4627 cfs_spin_lock_init(&o->od_osfs_lock);
4628 cfs_mutex_init(&o->od_otable_mutex);
4629 o->od_osfs_age = cfs_time_shift_64(-1000);
4631 o->od_capa_hash = init_capa_hash();
4632 if (o->od_capa_hash == NULL)
4633 GOTO(out, rc = -ENOMEM);
4635 o->od_read_cache = 1;
4636 o->od_writethrough_cache = 1;
4638 rc = osd_mount(env, o, cfg);
4642 /* setup scrub, including OI files initialization */
4643 rc = osd_scrub_setup(env, o);
4647 strncpy(o->od_svname, lustre_cfg_string(cfg, 4),
4648 sizeof(o->od_svname) - 1);
4650 if (strstr(o->od_svname, "-OST")) {
4651 rc = osd_compat_init(o);
4656 rc = lu_site_init(&o->od_site, l);
4658 GOTO(out_compat, rc);
4660 rc = lu_site_init_finish(&o->od_site);
4662 GOTO(out_compat, rc);
4664 rc = osd_procfs_init(o, o->od_svname);
4666 CERROR("%s: can't initialize procfs: rc = %d\n",
4668 GOTO(out_compat, rc);
4671 LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
4677 osd_oi_fini(info, o);
4678 osd_shutdown(env, o);
4682 cleanup_capa_hash(o->od_capa_hash);
4687 static struct lu_device *osd_device_alloc(const struct lu_env *env,
4688 struct lu_device_type *t,
4689 struct lustre_cfg *cfg)
4691 struct osd_device *o;
4696 return ERR_PTR(-ENOMEM);
4698 rc = dt_device_init(&o->od_dt_dev, t);
4700 rc = osd_device_init0(env, o, cfg);
4702 dt_device_fini(&o->od_dt_dev);
4705 if (unlikely(rc != 0))
4708 return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
4711 static struct lu_device *osd_device_free(const struct lu_env *env,
4712 struct lu_device *d)
4714 struct osd_device *o = osd_dev(d);
4717 cleanup_capa_hash(o->od_capa_hash);
4718 /* XXX: make osd top device in order to release reference */
4719 d->ld_site->ls_top_dev = d;
4720 lu_site_purge(env, d->ld_site, -1);
4721 if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
4722 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
4723 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
4725 lu_site_fini(&o->od_site);
4726 dt_device_fini(&o->od_dt_dev);
4731 static int osd_process_config(const struct lu_env *env,
4732 struct lu_device *d, struct lustre_cfg *cfg)
4734 struct osd_device *o = osd_dev(d);
4738 switch(cfg->lcfg_command) {
4740 err = osd_mount(env, o, cfg);
4743 lu_dev_del_linkage(d->ld_site, d);
4753 static int osd_recovery_complete(const struct lu_env *env,
4754 struct lu_device *d)
4760 * we use exports to track all osd users
4762 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
4763 struct obd_device *obd, struct obd_uuid *cluuid,
4764 struct obd_connect_data *data, void *localdata)
4766 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
4767 struct lustre_handle conn;
4771 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
4773 rc = class_connect(&conn, obd, cluuid);
4777 *exp = class_conn2export(&conn);
4779 cfs_spin_lock(&osd->od_osfs_lock);
4781 cfs_spin_unlock(&osd->od_osfs_lock);
4787 * once last export (we don't count self-export) disappeared
4788 * osd can be released
4790 static int osd_obd_disconnect(struct obd_export *exp)
4792 struct obd_device *obd = exp->exp_obd;
4793 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
4794 int rc, release = 0;
4797 /* Only disconnect the underlying layers on the final disconnect. */
4798 cfs_spin_lock(&osd->od_osfs_lock);
4800 if (osd->od_connects == 0)
4802 cfs_spin_unlock(&osd->od_osfs_lock);
4804 rc = class_disconnect(exp); /* bz 9811 */
4806 if (rc == 0 && release)
4807 class_manual_cleanup(obd);
4811 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
4812 struct lu_device *dev)
4814 struct osd_device *osd = osd_dev(dev);
4818 /* 2. setup quota slave instance */
4819 osd->od_quota_slave = qsd_init(env, osd->od_svname, &osd->od_dt_dev,
4820 osd->od_proc_entry);
4821 if (IS_ERR(osd->od_quota_slave)) {
4822 result = PTR_ERR(osd->od_quota_slave);
4823 osd->od_quota_slave = NULL;
4827 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 3, 55, 0)
4828 /* Unfortunately, the current MDD implementation relies on some specific
4829 * code to be executed in the OSD layer. Since OFD now also uses the OSD
4830 * module, we need a way to skip the metadata-specific code when running
4832 * The hack here is to check the type of the parent device which is
4833 * either MD (i.e. MDD device) with the current MDT stack or DT (i.e.
4834 * OFD device) on an OST. As a reminder, obdfilter does not use the OSD
4835 * layer and still relies on lvfs. This hack won't work any more when
4836 * LOD is landed since LOD is of DT type.
4837 * This code should be removed once the orion MDT changes (LOD/OSP, ...)
4838 * have been landed */
4839 osd->od_is_md = lu_device_is_md(pdev);
4841 #warning "all is_md checks must be removed from osd-ldiskfs"
4847 /* 3. setup local objects */
4848 result = llo_local_objects_setup(env, lu2md_dev(pdev), lu2dt_dev(dev));
4852 static const struct lu_object_operations osd_lu_obj_ops = {
4853 .loo_object_init = osd_object_init,
4854 .loo_object_delete = osd_object_delete,
4855 .loo_object_release = osd_object_release,
4856 .loo_object_free = osd_object_free,
4857 .loo_object_print = osd_object_print,
4858 .loo_object_invariant = osd_object_invariant
4861 const struct lu_device_operations osd_lu_ops = {
4862 .ldo_object_alloc = osd_object_alloc,
4863 .ldo_process_config = osd_process_config,
4864 .ldo_recovery_complete = osd_recovery_complete,
4865 .ldo_prepare = osd_prepare,
4868 static const struct lu_device_type_operations osd_device_type_ops = {
4869 .ldto_init = osd_type_init,
4870 .ldto_fini = osd_type_fini,
4872 .ldto_start = osd_type_start,
4873 .ldto_stop = osd_type_stop,
4875 .ldto_device_alloc = osd_device_alloc,
4876 .ldto_device_free = osd_device_free,
4878 .ldto_device_init = osd_device_init,
4879 .ldto_device_fini = osd_device_fini
4882 struct lu_device_type osd_device_type = {
4883 .ldt_tags = LU_DEVICE_DT,
4884 .ldt_name = LUSTRE_OSD_LDISKFS_NAME,
4885 .ldt_ops = &osd_device_type_ops,
4886 .ldt_ctx_tags = LCT_LOCAL,
4890 * lprocfs legacy support.
4892 static struct obd_ops osd_obd_device_ops = {
4893 .o_owner = THIS_MODULE,
4894 .o_connect = osd_obd_connect,
4895 .o_disconnect = osd_obd_disconnect
4898 static int __init osd_mod_init(void)
4900 struct lprocfs_static_vars lvars;
4903 lprocfs_osd_init_vars(&lvars);
4904 return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
4905 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
4908 static void __exit osd_mod_exit(void)
4910 class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
4913 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
4914 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
4915 MODULE_LICENSE("GPL");
4917 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);